executorch 1.1.0__cp313-cp313-win_amd64.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2784) hide show
  1. executorch/backends/__init__.py +0 -0
  2. executorch/backends/aoti/aoti_backend.py +271 -0
  3. executorch/backends/aoti/aoti_partitioner.py +109 -0
  4. executorch/backends/aoti/passes/replace_view_copy_with_view.py +118 -0
  5. executorch/backends/apple/coreml/compiler/__init__.py +9 -0
  6. executorch/backends/apple/coreml/compiler/coreml_preprocess.py +588 -0
  7. executorch/backends/apple/coreml/compiler/enumerated_shape_utils.py +244 -0
  8. executorch/backends/apple/coreml/compiler/torch_ops.py +219 -0
  9. executorch/backends/apple/coreml/executorchcoreml.pyi +10 -0
  10. executorch/backends/apple/coreml/logging.py +24 -0
  11. executorch/backends/apple/coreml/partition/__init__.py +9 -0
  12. executorch/backends/apple/coreml/partition/coreml_partitioner.py +307 -0
  13. executorch/backends/apple/coreml/quantizer/__init__.py +9 -0
  14. executorch/backends/apple/coreml/quantizer/_annotation_config.py +130 -0
  15. executorch/backends/apple/coreml/quantizer/_coreml_quantizer.py +628 -0
  16. executorch/backends/apple/coreml/quantizer/_coreml_quantizer_utils.py +800 -0
  17. executorch/backends/apple/coreml/quantizer/coreml_quantizer.py +7 -0
  18. executorch/backends/apple/coreml/recipes/__init__.py +17 -0
  19. executorch/backends/apple/coreml/recipes/coreml_recipe_provider.py +413 -0
  20. executorch/backends/apple/coreml/recipes/coreml_recipe_types.py +53 -0
  21. executorch/backends/apple/coreml/runtime/test/export_stateful_model.py +77 -0
  22. executorch/backends/apple/coreml/test/test_coreml_partitioner.py +348 -0
  23. executorch/backends/apple/coreml/test/test_coreml_quantizer.py +172 -0
  24. executorch/backends/apple/coreml/test/test_coreml_recipes.py +552 -0
  25. executorch/backends/apple/coreml/test/test_coreml_utils.py +19 -0
  26. executorch/backends/apple/coreml/test/test_enumerated_shapes.py +112 -0
  27. executorch/backends/apple/coreml/test/test_torch_ops.py +333 -0
  28. executorch/backends/apple/coreml/test/tester.py +121 -0
  29. executorch/backends/apple/metal/metal_backend.py +67 -0
  30. executorch/backends/apple/metal/metal_partitioner.py +25 -0
  31. executorch/backends/apple/metal/tests/__init__.py +5 -0
  32. executorch/backends/apple/metal/tests/test_metal_backend.py +79 -0
  33. executorch/backends/apple/metal/tests/test_metal_partitioner.py +172 -0
  34. executorch/backends/apple/mps/__init__.py +10 -0
  35. executorch/backends/apple/mps/mps_preprocess.py +301 -0
  36. executorch/backends/apple/mps/operators/__init__.py +77 -0
  37. executorch/backends/apple/mps/operators/activation_ops.py +99 -0
  38. executorch/backends/apple/mps/operators/binary_ops.py +156 -0
  39. executorch/backends/apple/mps/operators/clamp_ops.py +60 -0
  40. executorch/backends/apple/mps/operators/constant_ops.py +128 -0
  41. executorch/backends/apple/mps/operators/convolution_ops.py +70 -0
  42. executorch/backends/apple/mps/operators/indexing_ops.py +225 -0
  43. executorch/backends/apple/mps/operators/linear_algebra_ops.py +52 -0
  44. executorch/backends/apple/mps/operators/node_visitor.py +473 -0
  45. executorch/backends/apple/mps/operators/normalization_ops.py +97 -0
  46. executorch/backends/apple/mps/operators/op_clone.py +54 -0
  47. executorch/backends/apple/mps/operators/op_getitem.py +29 -0
  48. executorch/backends/apple/mps/operators/op_quant_dequant.py +142 -0
  49. executorch/backends/apple/mps/operators/op_skip_ops.py +24 -0
  50. executorch/backends/apple/mps/operators/pad_ops.py +37 -0
  51. executorch/backends/apple/mps/operators/pooling_ops.py +142 -0
  52. executorch/backends/apple/mps/operators/range_ops.py +53 -0
  53. executorch/backends/apple/mps/operators/reduce_ops.py +39 -0
  54. executorch/backends/apple/mps/operators/shape_ops.py +263 -0
  55. executorch/backends/apple/mps/operators/unary_ops.py +130 -0
  56. executorch/backends/apple/mps/partition/__init__.py +10 -0
  57. executorch/backends/apple/mps/partition/mps_partitioner.py +133 -0
  58. executorch/backends/apple/mps/serialization/mps_graph_schema.py +817 -0
  59. executorch/backends/apple/mps/serialization/mps_graph_serialize.py +34 -0
  60. executorch/backends/apple/mps/serialization/schema.fbs +523 -0
  61. executorch/backends/apple/mps/test/test_mps.py +2507 -0
  62. executorch/backends/apple/mps/test/test_mps_binary_ops.py +296 -0
  63. executorch/backends/apple/mps/test/test_mps_indexing_ops.py +264 -0
  64. executorch/backends/apple/mps/test/test_mps_linear.py +523 -0
  65. executorch/backends/apple/mps/test/test_mps_models.py +416 -0
  66. executorch/backends/apple/mps/test/test_mps_unary_ops.py +26 -0
  67. executorch/backends/apple/mps/test/test_mps_utils.py +326 -0
  68. executorch/backends/apple/mps/utils/mps_utils.py +82 -0
  69. executorch/backends/apple/mps/utils/quant_utils.py +41 -0
  70. executorch/backends/arm/_passes/__init__.py +134 -0
  71. executorch/backends/arm/_passes/_debug_passes.py +58 -0
  72. executorch/backends/arm/_passes/annotate_decomposed_matmul.py +127 -0
  73. executorch/backends/arm/_passes/annotate_output_dim_order_pass.py +26 -0
  74. executorch/backends/arm/_passes/arm_pass.py +95 -0
  75. executorch/backends/arm/_passes/arm_pass_manager.py +462 -0
  76. executorch/backends/arm/_passes/arm_pass_utils.py +281 -0
  77. executorch/backends/arm/_passes/broadcast_args_pass.py +72 -0
  78. executorch/backends/arm/_passes/cast_int64_pass.py +70 -0
  79. executorch/backends/arm/_passes/cast_to_int32_pass.py +67 -0
  80. executorch/backends/arm/_passes/conv1d_unsqueeze_pass.py +98 -0
  81. executorch/backends/arm/_passes/convert_elu_params.py +67 -0
  82. executorch/backends/arm/_passes/convert_expand_copy_to_repeat.py +77 -0
  83. executorch/backends/arm/_passes/convert_full_like_to_full_pass.py +42 -0
  84. executorch/backends/arm/_passes/convert_int64_const_ops_to_int32.py +81 -0
  85. executorch/backends/arm/_passes/convert_int64_output_ops_to_int32.py +155 -0
  86. executorch/backends/arm/_passes/convert_minmax_pass.py +166 -0
  87. executorch/backends/arm/_passes/convert_permute_singleton_to_view_pass.py +64 -0
  88. executorch/backends/arm/_passes/convert_split_to_slice.py +132 -0
  89. executorch/backends/arm/_passes/convert_squeezes_to_view.py +36 -0
  90. executorch/backends/arm/_passes/convert_to_clamp_pass.py +45 -0
  91. executorch/backends/arm/_passes/decompose_acosh_pass.py +77 -0
  92. executorch/backends/arm/_passes/decompose_adaptive_avg_pool2d_pass.py +112 -0
  93. executorch/backends/arm/_passes/decompose_add_sub_alpha_pass.py +98 -0
  94. executorch/backends/arm/_passes/decompose_addmm_pass.py +72 -0
  95. executorch/backends/arm/_passes/decompose_any_pass.py +114 -0
  96. executorch/backends/arm/_passes/decompose_asin_and_acos_pass.py +236 -0
  97. executorch/backends/arm/_passes/decompose_asinh_pass.py +74 -0
  98. executorch/backends/arm/_passes/decompose_atan_pass.py +142 -0
  99. executorch/backends/arm/_passes/decompose_atanh_pass.py +86 -0
  100. executorch/backends/arm/_passes/decompose_avg_pool2d_pass.py +154 -0
  101. executorch/backends/arm/_passes/decompose_batch_norm_no_stats.py +233 -0
  102. executorch/backends/arm/_passes/decompose_cosh_pass.py +72 -0
  103. executorch/backends/arm/_passes/decompose_cosine_similarity_pass.py +92 -0
  104. executorch/backends/arm/_passes/decompose_cumsum_pass.py +167 -0
  105. executorch/backends/arm/_passes/decompose_div_pass.py +61 -0
  106. executorch/backends/arm/_passes/decompose_div_tensor_mode.py +94 -0
  107. executorch/backends/arm/_passes/decompose_elu_pass.py +107 -0
  108. executorch/backends/arm/_passes/decompose_embedding_pass.py +126 -0
  109. executorch/backends/arm/_passes/decompose_expm1_pass.py +163 -0
  110. executorch/backends/arm/_passes/decompose_floor_divide_pass.py +75 -0
  111. executorch/backends/arm/_passes/decompose_gelu_pass.py +172 -0
  112. executorch/backends/arm/_passes/decompose_glu_pass.py +81 -0
  113. executorch/backends/arm/_passes/decompose_grouped_conv_pass.py +185 -0
  114. executorch/backends/arm/_passes/decompose_groupnorm_pass.py +227 -0
  115. executorch/backends/arm/_passes/decompose_int16_activation_conv_pass.py +137 -0
  116. executorch/backends/arm/_passes/decompose_int_pow_pass.py +64 -0
  117. executorch/backends/arm/_passes/decompose_layernorm_pass.py +203 -0
  118. executorch/backends/arm/_passes/decompose_leaky_relu_pass.py +76 -0
  119. executorch/backends/arm/_passes/decompose_linalg_vector_norm_pass.py +92 -0
  120. executorch/backends/arm/_passes/decompose_linear_pass.py +109 -0
  121. executorch/backends/arm/_passes/decompose_log1p_pass.py +97 -0
  122. executorch/backends/arm/_passes/decompose_logit_pass.py +119 -0
  123. executorch/backends/arm/_passes/decompose_masked_fill_pass.py +58 -0
  124. executorch/backends/arm/_passes/decompose_maxpool2d_with_dilation_pass.py +218 -0
  125. executorch/backends/arm/_passes/decompose_meandim_pass.py +283 -0
  126. executorch/backends/arm/_passes/decompose_ne_pass.py +74 -0
  127. executorch/backends/arm/_passes/decompose_quant_nodes.py +156 -0
  128. executorch/backends/arm/_passes/decompose_remainder_pass.py +70 -0
  129. executorch/backends/arm/_passes/decompose_round_pass.py +96 -0
  130. executorch/backends/arm/_passes/decompose_sdpa_pass.py +36 -0
  131. executorch/backends/arm/_passes/decompose_select.py +75 -0
  132. executorch/backends/arm/_passes/decompose_select_scatter_pass.py +143 -0
  133. executorch/backends/arm/_passes/decompose_sign_pass.py +78 -0
  134. executorch/backends/arm/_passes/decompose_silu_pass.py +41 -0
  135. executorch/backends/arm/_passes/decompose_sinh_pass.py +77 -0
  136. executorch/backends/arm/_passes/decompose_softmax_pass.py +93 -0
  137. executorch/backends/arm/_passes/decompose_softmax_unstable_pass.py +86 -0
  138. executorch/backends/arm/_passes/decompose_sqrt_pass.py +53 -0
  139. executorch/backends/arm/_passes/decompose_sum_pass.py +93 -0
  140. executorch/backends/arm/_passes/decompose_tan_pass.py +51 -0
  141. executorch/backends/arm/_passes/decompose_tosa_unsupported_clamp_pass.py +97 -0
  142. executorch/backends/arm/_passes/decompose_var_pass.py +109 -0
  143. executorch/backends/arm/_passes/decorate_fp32_to_int32_casting_pass.py +88 -0
  144. executorch/backends/arm/_passes/fold_qdq_with_annotated_qparams_pass.py +394 -0
  145. executorch/backends/arm/_passes/fuse_batch_norm2d_pass.py +241 -0
  146. executorch/backends/arm/_passes/fuse_constant_ops_pass.py +272 -0
  147. executorch/backends/arm/_passes/fuse_duplicate_users_pass.py +165 -0
  148. executorch/backends/arm/_passes/fuse_equal_placeholders_pass.py +104 -0
  149. executorch/backends/arm/_passes/fuse_quantized_activation_pass.py +80 -0
  150. executorch/backends/arm/_passes/fuse_view_copy_transform_pass.py +14 -0
  151. executorch/backends/arm/_passes/insert_int32_casts_after_int64_placeholders.py +130 -0
  152. executorch/backends/arm/_passes/insert_rescales_pass.py +584 -0
  153. executorch/backends/arm/_passes/insert_table_ops.py +305 -0
  154. executorch/backends/arm/_passes/match_arg_dtype_pass.py +104 -0
  155. executorch/backends/arm/_passes/match_arg_ranks_pass.py +114 -0
  156. executorch/backends/arm/_passes/mm_to_bmm_pass.py +104 -0
  157. executorch/backends/arm/_passes/normalize_while_initial_args_pass.py +113 -0
  158. executorch/backends/arm/_passes/promote_bool_operands_pass.py +88 -0
  159. executorch/backends/arm/_passes/quant_args.py +125 -0
  160. executorch/backends/arm/_passes/remove_getitem_pass.py +14 -0
  161. executorch/backends/arm/_passes/remove_graph_asserts_pass.py +18 -0
  162. executorch/backends/arm/_passes/remove_noop_pass.py +40 -0
  163. executorch/backends/arm/_passes/replace_inf_and_limit_values_pass.py +73 -0
  164. executorch/backends/arm/_passes/replace_scalar_with_tensor_pass.py +121 -0
  165. executorch/backends/arm/_passes/rewrite_bool_bitwise_not_to_logical_not_pass.py +43 -0
  166. executorch/backends/arm/_passes/rewrite_bool_to_fp32_cast_via_int8_pass.py +77 -0
  167. executorch/backends/arm/_passes/rewrite_conv_pass.py +336 -0
  168. executorch/backends/arm/_passes/rewrite_matmul.py +98 -0
  169. executorch/backends/arm/_passes/rewrite_upsample.py +93 -0
  170. executorch/backends/arm/_passes/scalars_to_attribute_pass.py +97 -0
  171. executorch/backends/arm/_passes/size_adjust_input_pass.py +227 -0
  172. executorch/backends/arm/_passes/to_tosa_memory_format_pass.py +454 -0
  173. executorch/backends/arm/_passes/unsqueeze_before_repeat_pass.py +66 -0
  174. executorch/backends/arm/_passes/unsqueeze_scalar_placeholders_pass.py +85 -0
  175. executorch/backends/arm/arm_vela.py +137 -0
  176. executorch/backends/arm/common/__init__.py +4 -0
  177. executorch/backends/arm/common/annotation_meta.py +39 -0
  178. executorch/backends/arm/common/arm_compile_spec.py +271 -0
  179. executorch/backends/arm/common/debug.py +79 -0
  180. executorch/backends/arm/common/pipeline_config.py +59 -0
  181. executorch/backends/arm/common/type.py +28 -0
  182. executorch/backends/arm/constants.py +48 -0
  183. executorch/backends/arm/debug/__init__.py +4 -0
  184. executorch/backends/arm/debug/schema.py +145 -0
  185. executorch/backends/arm/ethosu/__init__.py +11 -0
  186. executorch/backends/arm/ethosu/backend.py +115 -0
  187. executorch/backends/arm/ethosu/compile_spec.py +117 -0
  188. executorch/backends/arm/ethosu/partitioner.py +35 -0
  189. executorch/backends/arm/operator_support/__init__.py +23 -0
  190. executorch/backends/arm/operator_support/clone_dim_order_support.py +82 -0
  191. executorch/backends/arm/operator_support/control_flow_support.py +162 -0
  192. executorch/backends/arm/operator_support/convolution_support.py +160 -0
  193. executorch/backends/arm/operator_support/embedding_support.py +61 -0
  194. executorch/backends/arm/operator_support/ethos_u55_support.py +598 -0
  195. executorch/backends/arm/operator_support/index_select_support.py +60 -0
  196. executorch/backends/arm/operator_support/index_tensor_support.py +160 -0
  197. executorch/backends/arm/operator_support/minmax_support.py +61 -0
  198. executorch/backends/arm/operator_support/pool_2d_support.py +214 -0
  199. executorch/backends/arm/operator_support/reduce_sum_support.py +76 -0
  200. executorch/backends/arm/operator_support/right_shift_support.py +52 -0
  201. executorch/backends/arm/operator_support/slice_copy_support.py +48 -0
  202. executorch/backends/arm/operator_support/to_dim_order_copy_support.py +200 -0
  203. executorch/backends/arm/operator_support/tosa_profile_supported_op_lists.py +251 -0
  204. executorch/backends/arm/operator_support/tosa_supported_operators.py +748 -0
  205. executorch/backends/arm/operator_support/where_support.py +77 -0
  206. executorch/backends/arm/operators/__init__.py +71 -0
  207. executorch/backends/arm/operators/node_visitor.py +177 -0
  208. executorch/backends/arm/operators/op_abs.py +59 -0
  209. executorch/backends/arm/operators/op_add.py +60 -0
  210. executorch/backends/arm/operators/op_amax.py +72 -0
  211. executorch/backends/arm/operators/op_amin.py +73 -0
  212. executorch/backends/arm/operators/op_any.py +61 -0
  213. executorch/backends/arm/operators/op_avg_pool2d.py +142 -0
  214. executorch/backends/arm/operators/op_bitwise_not.py +62 -0
  215. executorch/backends/arm/operators/op_cat.py +68 -0
  216. executorch/backends/arm/operators/op_ceil.py +56 -0
  217. executorch/backends/arm/operators/op_clamp.py +118 -0
  218. executorch/backends/arm/operators/op_cond_if.py +56 -0
  219. executorch/backends/arm/operators/op_constant_pad_nd.py +117 -0
  220. executorch/backends/arm/operators/op_cos.py +49 -0
  221. executorch/backends/arm/operators/op_eq.py +64 -0
  222. executorch/backends/arm/operators/op_erf.py +54 -0
  223. executorch/backends/arm/operators/op_exp.py +54 -0
  224. executorch/backends/arm/operators/op_floor.py +56 -0
  225. executorch/backends/arm/operators/op_ge.py +64 -0
  226. executorch/backends/arm/operators/op_gt.py +64 -0
  227. executorch/backends/arm/operators/op_index_select.py +109 -0
  228. executorch/backends/arm/operators/op_index_tensor.py +247 -0
  229. executorch/backends/arm/operators/op_le.py +64 -0
  230. executorch/backends/arm/operators/op_log.py +50 -0
  231. executorch/backends/arm/operators/op_logical_not.py +62 -0
  232. executorch/backends/arm/operators/op_lt.py +64 -0
  233. executorch/backends/arm/operators/op_max_pool2d.py +108 -0
  234. executorch/backends/arm/operators/op_maximum.py +66 -0
  235. executorch/backends/arm/operators/op_minimum.py +67 -0
  236. executorch/backends/arm/operators/op_mul.py +61 -0
  237. executorch/backends/arm/operators/op_neg.py +92 -0
  238. executorch/backends/arm/operators/op_permute.py +150 -0
  239. executorch/backends/arm/operators/op_pow.py +63 -0
  240. executorch/backends/arm/operators/op_reciprocal.py +51 -0
  241. executorch/backends/arm/operators/op_repeat.py +77 -0
  242. executorch/backends/arm/operators/op_rshift_tensor.py +62 -0
  243. executorch/backends/arm/operators/op_rsqrt.py +51 -0
  244. executorch/backends/arm/operators/op_sigmoid.py +50 -0
  245. executorch/backends/arm/operators/op_sin.py +49 -0
  246. executorch/backends/arm/operators/op_slice.py +151 -0
  247. executorch/backends/arm/operators/op_sub.py +63 -0
  248. executorch/backends/arm/operators/op_sum.py +64 -0
  249. executorch/backends/arm/operators/op_tanh.py +51 -0
  250. executorch/backends/arm/operators/op_to_dim_order_copy.py +49 -0
  251. executorch/backends/arm/operators/op_tosa_conv2d.py +139 -0
  252. executorch/backends/arm/operators/op_tosa_conv3d.py +24 -0
  253. executorch/backends/arm/operators/op_tosa_depthwise_conv2d.py +32 -0
  254. executorch/backends/arm/operators/op_tosa_matmul.py +102 -0
  255. executorch/backends/arm/operators/op_tosa_rescale.py +261 -0
  256. executorch/backends/arm/operators/op_tosa_resize.py +124 -0
  257. executorch/backends/arm/operators/op_tosa_table.py +72 -0
  258. executorch/backends/arm/operators/op_tosa_transpose.py +71 -0
  259. executorch/backends/arm/operators/op_view.py +81 -0
  260. executorch/backends/arm/operators/op_where.py +75 -0
  261. executorch/backends/arm/operators/op_while.py +103 -0
  262. executorch/backends/arm/operators/operator_validation_utils.py +201 -0
  263. executorch/backends/arm/operators/ops_binary.py +112 -0
  264. executorch/backends/arm/operators/ops_identity.py +80 -0
  265. executorch/backends/arm/process_node.py +230 -0
  266. executorch/backends/arm/quantizer/__init__.py +35 -0
  267. executorch/backends/arm/quantizer/arm_quantizer.py +710 -0
  268. executorch/backends/arm/quantizer/arm_quantizer_utils.py +76 -0
  269. executorch/backends/arm/quantizer/quantization_annotator.py +815 -0
  270. executorch/backends/arm/quantizer/quantization_config.py +228 -0
  271. executorch/backends/arm/scripts/collect_testname_resources.py +194 -0
  272. executorch/backends/arm/scripts/parse_test_names.py +271 -0
  273. executorch/backends/arm/test/common.py +302 -0
  274. executorch/backends/arm/test/conftest.py +134 -0
  275. executorch/backends/arm/test/misc/test_bn_relu_folding_qat.py +124 -0
  276. executorch/backends/arm/test/misc/test_call_operator_submodule.py +72 -0
  277. executorch/backends/arm/test/misc/test_compile_spec.py +55 -0
  278. executorch/backends/arm/test/misc/test_conv_relu_residual_add.py +108 -0
  279. executorch/backends/arm/test/misc/test_custom_partition.py +183 -0
  280. executorch/backends/arm/test/misc/test_debug_feats.py +316 -0
  281. executorch/backends/arm/test/misc/test_debug_hook.py +186 -0
  282. executorch/backends/arm/test/misc/test_dim_order.py +127 -0
  283. executorch/backends/arm/test/misc/test_dw_convs_with_shared_weights.py +46 -0
  284. executorch/backends/arm/test/misc/test_extract_io_params_tosa.py +92 -0
  285. executorch/backends/arm/test/misc/test_int64.py +112 -0
  286. executorch/backends/arm/test/misc/test_lifted_tensor.py +146 -0
  287. executorch/backends/arm/test/misc/test_mixed_type_lowering.py +83 -0
  288. executorch/backends/arm/test/misc/test_model_evaluator.py +58 -0
  289. executorch/backends/arm/test/misc/test_multiple_delegates.py +53 -0
  290. executorch/backends/arm/test/misc/test_multiple_outputs.py +70 -0
  291. executorch/backends/arm/test/misc/test_non_persistent_buffers.py +49 -0
  292. executorch/backends/arm/test/misc/test_outputs_order.py +122 -0
  293. executorch/backends/arm/test/misc/test_partition_decomposed_quantized_ops.py +181 -0
  294. executorch/backends/arm/test/misc/test_pass_pipeline_config.py +35 -0
  295. executorch/backends/arm/test/misc/test_pass_required_order.py +97 -0
  296. executorch/backends/arm/test/misc/test_qat_training_loop.py +100 -0
  297. executorch/backends/arm/test/misc/test_quant_custom_meta.py +105 -0
  298. executorch/backends/arm/test/misc/test_rescale_range.py +130 -0
  299. executorch/backends/arm/test/misc/test_save_exported_model.py +64 -0
  300. executorch/backends/arm/test/misc/test_tosa_dialect_conv2d.py +220 -0
  301. executorch/backends/arm/test/misc/test_tosa_dialect_dw_conv2d.py +227 -0
  302. executorch/backends/arm/test/misc/test_tosa_spec.py +169 -0
  303. executorch/backends/arm/test/models/stable_diffusion/stable_diffusion_module_test_configs.py +114 -0
  304. executorch/backends/arm/test/models/stable_diffusion/test_CLIPTextModelWithProjection.py +168 -0
  305. executorch/backends/arm/test/models/stable_diffusion/test_SD3Transformer2DModel.py +193 -0
  306. executorch/backends/arm/test/models/stable_diffusion/test_T5EncoderModel.py +161 -0
  307. executorch/backends/arm/test/models/stable_diffusion/test_vae_AutoencoderKL.py +118 -0
  308. executorch/backends/arm/test/models/test_conformer.py +138 -0
  309. executorch/backends/arm/test/models/test_deit_tiny_arm.py +120 -0
  310. executorch/backends/arm/test/models/test_dl3_arm.py +118 -0
  311. executorch/backends/arm/test/models/test_inception_v3_arm.py +120 -0
  312. executorch/backends/arm/test/models/test_llama.py +227 -0
  313. executorch/backends/arm/test/models/test_lstm_arm.py +194 -0
  314. executorch/backends/arm/test/models/test_mobilenet_v2_arm.py +179 -0
  315. executorch/backends/arm/test/models/test_mobilenet_v3_arm.py +113 -0
  316. executorch/backends/arm/test/models/test_nn_functional.py +124 -0
  317. executorch/backends/arm/test/models/test_nn_modules.py +161 -0
  318. executorch/backends/arm/test/models/test_nss.py +146 -0
  319. executorch/backends/arm/test/models/test_resnet18.py +98 -0
  320. executorch/backends/arm/test/models/test_torch_functions.py +151 -0
  321. executorch/backends/arm/test/models/test_w2l_arm.py +136 -0
  322. executorch/backends/arm/test/ops/test_abs.py +101 -0
  323. executorch/backends/arm/test/ops/test_acos.py +121 -0
  324. executorch/backends/arm/test/ops/test_acosh.py +137 -0
  325. executorch/backends/arm/test/ops/test_adaptive_avg_pool2d.py +242 -0
  326. executorch/backends/arm/test/ops/test_add.py +291 -0
  327. executorch/backends/arm/test/ops/test_addmm.py +285 -0
  328. executorch/backends/arm/test/ops/test_alias_copy.py +114 -0
  329. executorch/backends/arm/test/ops/test_amax.py +191 -0
  330. executorch/backends/arm/test/ops/test_amin.py +205 -0
  331. executorch/backends/arm/test/ops/test_any.py +210 -0
  332. executorch/backends/arm/test/ops/test_arange.py +258 -0
  333. executorch/backends/arm/test/ops/test_asin.py +105 -0
  334. executorch/backends/arm/test/ops/test_asinh.py +104 -0
  335. executorch/backends/arm/test/ops/test_at.py +256 -0
  336. executorch/backends/arm/test/ops/test_atan.py +111 -0
  337. executorch/backends/arm/test/ops/test_atanh.py +112 -0
  338. executorch/backends/arm/test/ops/test_avg_pool2d.py +295 -0
  339. executorch/backends/arm/test/ops/test_batch_norm.py +385 -0
  340. executorch/backends/arm/test/ops/test_bitwise.py +672 -0
  341. executorch/backends/arm/test/ops/test_bitwise_not.py +115 -0
  342. executorch/backends/arm/test/ops/test_bmm.py +193 -0
  343. executorch/backends/arm/test/ops/test_cat.py +219 -0
  344. executorch/backends/arm/test/ops/test_ceil.py +125 -0
  345. executorch/backends/arm/test/ops/test_clamp.py +474 -0
  346. executorch/backends/arm/test/ops/test_clone.py +150 -0
  347. executorch/backends/arm/test/ops/test_cond.py +338 -0
  348. executorch/backends/arm/test/ops/test_constant_pad_nd.py +119 -0
  349. executorch/backends/arm/test/ops/test_conv1d.py +356 -0
  350. executorch/backends/arm/test/ops/test_conv2d.py +592 -0
  351. executorch/backends/arm/test/ops/test_conv3d.py +711 -0
  352. executorch/backends/arm/test/ops/test_conv_combos.py +668 -0
  353. executorch/backends/arm/test/ops/test_conv_constant_pad_nd.py +145 -0
  354. executorch/backends/arm/test/ops/test_copy.py +171 -0
  355. executorch/backends/arm/test/ops/test_cos.py +115 -0
  356. executorch/backends/arm/test/ops/test_cosh.py +111 -0
  357. executorch/backends/arm/test/ops/test_cumsum.py +122 -0
  358. executorch/backends/arm/test/ops/test_depthwise_conv.py +404 -0
  359. executorch/backends/arm/test/ops/test_div.py +151 -0
  360. executorch/backends/arm/test/ops/test_div_tensor_mode.py +151 -0
  361. executorch/backends/arm/test/ops/test_elu.py +133 -0
  362. executorch/backends/arm/test/ops/test_embedding.py +146 -0
  363. executorch/backends/arm/test/ops/test_eq.py +286 -0
  364. executorch/backends/arm/test/ops/test_erf.py +96 -0
  365. executorch/backends/arm/test/ops/test_exp.py +110 -0
  366. executorch/backends/arm/test/ops/test_expand.py +153 -0
  367. executorch/backends/arm/test/ops/test_expm1.py +113 -0
  368. executorch/backends/arm/test/ops/test_eye.py +156 -0
  369. executorch/backends/arm/test/ops/test_fill_scalar.py +108 -0
  370. executorch/backends/arm/test/ops/test_floor.py +125 -0
  371. executorch/backends/arm/test/ops/test_floor_div.py +156 -0
  372. executorch/backends/arm/test/ops/test_full.py +255 -0
  373. executorch/backends/arm/test/ops/test_ge.py +292 -0
  374. executorch/backends/arm/test/ops/test_gelu.py +156 -0
  375. executorch/backends/arm/test/ops/test_glu.py +130 -0
  376. executorch/backends/arm/test/ops/test_group_norm.py +200 -0
  377. executorch/backends/arm/test/ops/test_gt.py +293 -0
  378. executorch/backends/arm/test/ops/test_hardsigmoid.py +114 -0
  379. executorch/backends/arm/test/ops/test_hardswish.py +104 -0
  380. executorch/backends/arm/test/ops/test_hardtanh.py +113 -0
  381. executorch/backends/arm/test/ops/test_index_select.py +177 -0
  382. executorch/backends/arm/test/ops/test_index_tensor.py +467 -0
  383. executorch/backends/arm/test/ops/test_layer_norm.py +186 -0
  384. executorch/backends/arm/test/ops/test_le.py +296 -0
  385. executorch/backends/arm/test/ops/test_leaky_relu.py +125 -0
  386. executorch/backends/arm/test/ops/test_linalg_vector_norm.py +160 -0
  387. executorch/backends/arm/test/ops/test_linear.py +389 -0
  388. executorch/backends/arm/test/ops/test_log.py +100 -0
  389. executorch/backends/arm/test/ops/test_log1p.py +78 -0
  390. executorch/backends/arm/test/ops/test_logical.py +429 -0
  391. executorch/backends/arm/test/ops/test_logit.py +119 -0
  392. executorch/backends/arm/test/ops/test_logsoftmax.py +124 -0
  393. executorch/backends/arm/test/ops/test_lshift.py +220 -0
  394. executorch/backends/arm/test/ops/test_lt.py +293 -0
  395. executorch/backends/arm/test/ops/test_masked_fill.py +171 -0
  396. executorch/backends/arm/test/ops/test_matmul.py +289 -0
  397. executorch/backends/arm/test/ops/test_max_pool.py +329 -0
  398. executorch/backends/arm/test/ops/test_maximum.py +98 -0
  399. executorch/backends/arm/test/ops/test_mean_dim.py +388 -0
  400. executorch/backends/arm/test/ops/test_minimum.py +98 -0
  401. executorch/backends/arm/test/ops/test_mm.py +93 -0
  402. executorch/backends/arm/test/ops/test_mul.py +376 -0
  403. executorch/backends/arm/test/ops/test_multihead_attention.py +138 -0
  404. executorch/backends/arm/test/ops/test_ne.py +243 -0
  405. executorch/backends/arm/test/ops/test_neg.py +99 -0
  406. executorch/backends/arm/test/ops/test_ones.py +152 -0
  407. executorch/backends/arm/test/ops/test_permute.py +256 -0
  408. executorch/backends/arm/test/ops/test_pixel_shuffling.py +237 -0
  409. executorch/backends/arm/test/ops/test_pow.py +204 -0
  410. executorch/backends/arm/test/ops/test_reciprocal.py +112 -0
  411. executorch/backends/arm/test/ops/test_relu.py +166 -0
  412. executorch/backends/arm/test/ops/test_remainder.py +199 -0
  413. executorch/backends/arm/test/ops/test_repeat.py +162 -0
  414. executorch/backends/arm/test/ops/test_round.py +111 -0
  415. executorch/backends/arm/test/ops/test_rshift.py +226 -0
  416. executorch/backends/arm/test/ops/test_rsqrt.py +153 -0
  417. executorch/backends/arm/test/ops/test_rsub.py +126 -0
  418. executorch/backends/arm/test/ops/test_scalar_tensor.py +135 -0
  419. executorch/backends/arm/test/ops/test_scalars.py +438 -0
  420. executorch/backends/arm/test/ops/test_sdpa.py +73 -0
  421. executorch/backends/arm/test/ops/test_select.py +219 -0
  422. executorch/backends/arm/test/ops/test_select_scatter.py +173 -0
  423. executorch/backends/arm/test/ops/test_sigmoid.py +357 -0
  424. executorch/backends/arm/test/ops/test_sigmoid_32bit.py +177 -0
  425. executorch/backends/arm/test/ops/test_sign.py +113 -0
  426. executorch/backends/arm/test/ops/test_silu.py +174 -0
  427. executorch/backends/arm/test/ops/test_sin.py +108 -0
  428. executorch/backends/arm/test/ops/test_sinh.py +103 -0
  429. executorch/backends/arm/test/ops/test_slice.py +217 -0
  430. executorch/backends/arm/test/ops/test_softmax.py +121 -0
  431. executorch/backends/arm/test/ops/test_split.py +312 -0
  432. executorch/backends/arm/test/ops/test_sqrt.py +112 -0
  433. executorch/backends/arm/test/ops/test_squeeze.py +291 -0
  434. executorch/backends/arm/test/ops/test_stack.py +150 -0
  435. executorch/backends/arm/test/ops/test_sub.py +366 -0
  436. executorch/backends/arm/test/ops/test_sum.py +161 -0
  437. executorch/backends/arm/test/ops/test_t_copy.py +115 -0
  438. executorch/backends/arm/test/ops/test_tan.py +146 -0
  439. executorch/backends/arm/test/ops/test_tanh.py +169 -0
  440. executorch/backends/arm/test/ops/test_to_copy.py +296 -0
  441. executorch/backends/arm/test/ops/test_transpose_copy.py +114 -0
  442. executorch/backends/arm/test/ops/test_unary_combos.py +144 -0
  443. executorch/backends/arm/test/ops/test_unbind.py +82 -0
  444. executorch/backends/arm/test/ops/test_unflatten.py +107 -0
  445. executorch/backends/arm/test/ops/test_unsqueeze.py +107 -0
  446. executorch/backends/arm/test/ops/test_upsample_bilinear2d.py +443 -0
  447. executorch/backends/arm/test/ops/test_upsample_nearest2d.py +520 -0
  448. executorch/backends/arm/test/ops/test_var.py +394 -0
  449. executorch/backends/arm/test/ops/test_view.py +253 -0
  450. executorch/backends/arm/test/ops/test_where.py +285 -0
  451. executorch/backends/arm/test/ops/test_while.py +264 -0
  452. executorch/backends/arm/test/ops/test_zeros.py +158 -0
  453. executorch/backends/arm/test/passes/test_broadcast_args_pass.py +57 -0
  454. executorch/backends/arm/test/passes/test_cast_int64_pass.py +49 -0
  455. executorch/backends/arm/test/passes/test_convert_expand_copy_to_repeat.py +51 -0
  456. executorch/backends/arm/test/passes/test_convert_int64_const_ops_to_int32.py +529 -0
  457. executorch/backends/arm/test/passes/test_convert_int64_output_ops_to_int32.py +129 -0
  458. executorch/backends/arm/test/passes/test_convert_permute_singleton_to_view_pass.py +100 -0
  459. executorch/backends/arm/test/passes/test_convert_split_to_slice.py +75 -0
  460. executorch/backends/arm/test/passes/test_convert_to_clamp.py +92 -0
  461. executorch/backends/arm/test/passes/test_decompose_avg_pool2d_pass.py +82 -0
  462. executorch/backends/arm/test/passes/test_decompose_cosine_similarity_pass.py +58 -0
  463. executorch/backends/arm/test/passes/test_decompose_div_pass.py +70 -0
  464. executorch/backends/arm/test/passes/test_decompose_int_pow_pass.py +80 -0
  465. executorch/backends/arm/test/passes/test_decompose_layernorm_pass.py +69 -0
  466. executorch/backends/arm/test/passes/test_decompose_linalg_vector_norm_pass.py +107 -0
  467. executorch/backends/arm/test/passes/test_decompose_meandim_pass.py +137 -0
  468. executorch/backends/arm/test/passes/test_decompose_quant_nodes.py +44 -0
  469. executorch/backends/arm/test/passes/test_decompose_softmax_pass.py +103 -0
  470. executorch/backends/arm/test/passes/test_decompose_tosa_unsupported_clamp_pass.py +73 -0
  471. executorch/backends/arm/test/passes/test_decompose_var_pass.py +89 -0
  472. executorch/backends/arm/test/passes/test_decorate_fp32_to_int32_casting_pass.py +85 -0
  473. executorch/backends/arm/test/passes/test_fold_qdq_pass.py +51 -0
  474. executorch/backends/arm/test/passes/test_fuse_batchnorm_pass.py +164 -0
  475. executorch/backends/arm/test/passes/test_fuse_constant_ops_pass.py +177 -0
  476. executorch/backends/arm/test/passes/test_fuse_duplicate_users_pass.py +70 -0
  477. executorch/backends/arm/test/passes/test_fuse_equal_placeholders_ops_pass.py +170 -0
  478. executorch/backends/arm/test/passes/test_fuse_view_copy.py +82 -0
  479. executorch/backends/arm/test/passes/test_insert_int32_casts_after_int64_placeholders_pass.py +114 -0
  480. executorch/backends/arm/test/passes/test_insert_rescale_i32_pass.py +107 -0
  481. executorch/backends/arm/test/passes/test_insert_table_ops_pass.py +47 -0
  482. executorch/backends/arm/test/passes/test_int32_cast_embedding_pass.py +46 -0
  483. executorch/backends/arm/test/passes/test_ioquantization_pass.py +48 -0
  484. executorch/backends/arm/test/passes/test_promote_bool_operands_pass.py +103 -0
  485. executorch/backends/arm/test/passes/test_remove_clone_pass.py +45 -0
  486. executorch/backends/arm/test/passes/test_replace_inf_values_pass.py +96 -0
  487. executorch/backends/arm/test/passes/test_rescale_pass.py +81 -0
  488. executorch/backends/arm/test/passes/test_rewrite_conv_pass.py +20 -0
  489. executorch/backends/arm/test/passes/test_to_tosa_memory_format.py +213 -0
  490. executorch/backends/arm/test/passes/test_unsqueeze_before_repeat_pass.py +62 -0
  491. executorch/backends/arm/test/quantizer/test_conv_relu_fusing.py +118 -0
  492. executorch/backends/arm/test/quantizer/test_generic_annotater.py +109 -0
  493. executorch/backends/arm/test/quantizer/test_partial_quantization.py +104 -0
  494. executorch/backends/arm/test/quantizer/test_selective_quantization.py +212 -0
  495. executorch/backends/arm/test/quantizer/test_set_module_name.py +158 -0
  496. executorch/backends/arm/test/quantizer/test_tosa_quantizer_validate.py +55 -0
  497. executorch/backends/arm/test/runner_utils.py +840 -0
  498. executorch/backends/arm/test/test_memory_allocator_log.py +170 -0
  499. executorch/backends/arm/test/test_model.py +345 -0
  500. executorch/backends/arm/test/tester/__init__.py +6 -0
  501. executorch/backends/arm/test/tester/analyze_output_utils.py +346 -0
  502. executorch/backends/arm/test/tester/arm_tester.py +1149 -0
  503. executorch/backends/arm/test/tester/quantize.py +43 -0
  504. executorch/backends/arm/test/tester/serialize.py +78 -0
  505. executorch/backends/arm/test/tester/test_pipeline.py +1190 -0
  506. executorch/backends/arm/tosa/__init__.py +9 -0
  507. executorch/backends/arm/tosa/backend.py +443 -0
  508. executorch/backends/arm/tosa/compile_spec.py +48 -0
  509. executorch/backends/arm/tosa/dialect/__init__.py +15 -0
  510. executorch/backends/arm/tosa/dialect/lib.py +85 -0
  511. executorch/backends/arm/tosa/dialect/ops/conv2d.py +111 -0
  512. executorch/backends/arm/tosa/dialect/ops/conv3d.py +74 -0
  513. executorch/backends/arm/tosa/dialect/ops/depthwise_conv2d.py +56 -0
  514. executorch/backends/arm/tosa/dialect/ops/matmul.py +56 -0
  515. executorch/backends/arm/tosa/dialect/ops/rescale.py +53 -0
  516. executorch/backends/arm/tosa/dialect/ops/resize.py +66 -0
  517. executorch/backends/arm/tosa/dialect/ops/table.py +53 -0
  518. executorch/backends/arm/tosa/dialect/ops/transpose.py +35 -0
  519. executorch/backends/arm/tosa/dialect/ops_registration.py +68 -0
  520. executorch/backends/arm/tosa/mapping.py +309 -0
  521. executorch/backends/arm/tosa/partitioner.py +486 -0
  522. executorch/backends/arm/tosa/schemas/__init__.py +4 -0
  523. executorch/backends/arm/tosa/schemas/tosa_1.0.fbs +559 -0
  524. executorch/backends/arm/tosa/specification.py +406 -0
  525. executorch/backends/arm/tosa/utils.py +304 -0
  526. executorch/backends/arm/util/_factory.py +59 -0
  527. executorch/backends/arm/util/arm_model_evaluator.py +388 -0
  528. executorch/backends/arm/vgf/__init__.py +11 -0
  529. executorch/backends/arm/vgf/backend.py +172 -0
  530. executorch/backends/arm/vgf/compile_spec.py +73 -0
  531. executorch/backends/arm/vgf/model_converter.py +34 -0
  532. executorch/backends/arm/vgf/partitioner.py +35 -0
  533. executorch/backends/cadence/aot/__init__.py +5 -0
  534. executorch/backends/cadence/aot/compiler.py +484 -0
  535. executorch/backends/cadence/aot/compiler_funcs.py +203 -0
  536. executorch/backends/cadence/aot/compiler_utils.py +293 -0
  537. executorch/backends/cadence/aot/decompose_ops.py +120 -0
  538. executorch/backends/cadence/aot/export_example.py +117 -0
  539. executorch/backends/cadence/aot/functions.yaml +470 -0
  540. executorch/backends/cadence/aot/functions_fusion_g3.yaml +191 -0
  541. executorch/backends/cadence/aot/functions_hifi.yaml +565 -0
  542. executorch/backends/cadence/aot/functions_vision.yaml +275 -0
  543. executorch/backends/cadence/aot/fuse_ops.py +1138 -0
  544. executorch/backends/cadence/aot/graph_builder.py +142 -0
  545. executorch/backends/cadence/aot/memory_constraints.py +731 -0
  546. executorch/backends/cadence/aot/memory_planning.py +469 -0
  547. executorch/backends/cadence/aot/memory_planning_algo.py +341 -0
  548. executorch/backends/cadence/aot/ops_registrations.py +2841 -0
  549. executorch/backends/cadence/aot/pass_utils.py +274 -0
  550. executorch/backends/cadence/aot/passes.py +132 -0
  551. executorch/backends/cadence/aot/program_builder.py +138 -0
  552. executorch/backends/cadence/aot/quantizer/fusion_pass.py +743 -0
  553. executorch/backends/cadence/aot/quantizer/patterns.py +738 -0
  554. executorch/backends/cadence/aot/quantizer/quantizer.py +409 -0
  555. executorch/backends/cadence/aot/quantizer/utils.py +288 -0
  556. executorch/backends/cadence/aot/ref_implementations.py +2210 -0
  557. executorch/backends/cadence/aot/remove_ops.py +936 -0
  558. executorch/backends/cadence/aot/reorder_ops.py +843 -0
  559. executorch/backends/cadence/aot/replace_ops.py +2526 -0
  560. executorch/backends/cadence/aot/simplify_ops.py +221 -0
  561. executorch/backends/cadence/aot/tests/test_decompose_ops_passes.py +80 -0
  562. executorch/backends/cadence/aot/tests/test_fusion_ops_passes.py +1023 -0
  563. executorch/backends/cadence/aot/tests/test_graph_builder.py +104 -0
  564. executorch/backends/cadence/aot/tests/test_idma_ops.py +31 -0
  565. executorch/backends/cadence/aot/tests/test_memory_passes.py +1367 -0
  566. executorch/backends/cadence/aot/tests/test_pass_filter.py +167 -0
  567. executorch/backends/cadence/aot/tests/test_program_builder.py +222 -0
  568. executorch/backends/cadence/aot/tests/test_quantizer_ops.py +649 -0
  569. executorch/backends/cadence/aot/tests/test_ref_implementations.py +3151 -0
  570. executorch/backends/cadence/aot/tests/test_remove_ops_passes.py +977 -0
  571. executorch/backends/cadence/aot/tests/test_reorder_ops_passes.py +635 -0
  572. executorch/backends/cadence/aot/tests/test_replace_ops_passes.py +2777 -0
  573. executorch/backends/cadence/aot/tests/test_simplify_ops_passes.py +160 -0
  574. executorch/backends/cadence/aot/tests/test_type_dispatch_passes.py +481 -0
  575. executorch/backends/cadence/aot/type_dispatch.py +190 -0
  576. executorch/backends/cadence/aot/typing_stubs.py +22 -0
  577. executorch/backends/cadence/aot/utils.py +363 -0
  578. executorch/backends/cadence/runtime/__init__.py +13 -0
  579. executorch/backends/cadence/runtime/etdump.py +173 -0
  580. executorch/backends/cadence/runtime/executor.py +202 -0
  581. executorch/backends/cadence/runtime/runtime.py +156 -0
  582. executorch/backends/cadence/runtime/utils.py +100 -0
  583. executorch/backends/cadence/utils/FACTO/calibrator/__init__.py +0 -0
  584. executorch/backends/cadence/utils/FACTO/calibrator/runner.py +190 -0
  585. executorch/backends/cadence/utils/FACTO/examples/example.py +59 -0
  586. executorch/backends/cadence/utils/FACTO/examples/minimal_example.py +21 -0
  587. executorch/backends/cadence/utils/FACTO/examples/random_seed.py +35 -0
  588. executorch/backends/cadence/utils/FACTO/facto/__init__.py +0 -0
  589. executorch/backends/cadence/utils/FACTO/facto/inputgen/__init__.py +0 -0
  590. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/__init__.py +0 -0
  591. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/engine.py +129 -0
  592. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/gen.py +47 -0
  593. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/__init__.py +0 -0
  594. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/engine.py +291 -0
  595. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/gen.py +139 -0
  596. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/type.py +134 -0
  597. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/__init__.py +0 -0
  598. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/engine.py +54 -0
  599. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/model.py +133 -0
  600. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/solve.py +125 -0
  601. executorch/backends/cadence/utils/FACTO/facto/inputgen/specs/__init__.py +0 -0
  602. executorch/backends/cadence/utils/FACTO/facto/inputgen/specs/model.py +149 -0
  603. executorch/backends/cadence/utils/FACTO/facto/inputgen/utils/__init__.py +0 -0
  604. executorch/backends/cadence/utils/FACTO/facto/inputgen/utils/random_manager.py +33 -0
  605. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/__init__.py +0 -0
  606. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/constants.py +9 -0
  607. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/gen.py +283 -0
  608. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/solve.py +142 -0
  609. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/space.py +450 -0
  610. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/type.py +105 -0
  611. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/utils.py +30 -0
  612. executorch/backends/cadence/utils/FACTO/facto/specdb/db.py +4361 -0
  613. executorch/backends/cadence/utils/FACTO/facto/specdb/default.py +36 -0
  614. executorch/backends/cadence/utils/FACTO/facto/specdb/dtypes.py +25 -0
  615. executorch/backends/cadence/utils/FACTO/facto/specdb/function.py +537 -0
  616. executorch/backends/cadence/utils/FACTO/setup.py +19 -0
  617. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argtuple_engine.py +53 -0
  618. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argtuple_generator.py +58 -0
  619. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_attributes.py +78 -0
  620. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_generator.py +98 -0
  621. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_types.py +28 -0
  622. executorch/backends/cadence/utils/FACTO/test/inputgen/test_attribute_engine.py +62 -0
  623. executorch/backends/cadence/utils/FACTO/test/inputgen/test_attribute_solver.py +83 -0
  624. executorch/backends/cadence/utils/FACTO/test/inputgen/test_constraints.py +27 -0
  625. executorch/backends/cadence/utils/FACTO/test/inputgen/test_meta_arg_engine.py +70 -0
  626. executorch/backends/cadence/utils/FACTO/test/inputgen/test_specs.py +48 -0
  627. executorch/backends/cadence/utils/FACTO/test/inputgen/test_structural_engine.py +32 -0
  628. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_generation.py +80 -0
  629. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_solving.py +193 -0
  630. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_space.py +231 -0
  631. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_types.py +71 -0
  632. executorch/backends/cadence/utils/facto_util.py +611 -0
  633. executorch/backends/cadence/utils/gen_header.py +49 -0
  634. executorch/backends/cadence/utils/post_compilation.py +173 -0
  635. executorch/backends/cortex_m/ops/operators.py +890 -0
  636. executorch/backends/cortex_m/ops/operators.yaml +78 -0
  637. executorch/backends/cortex_m/passes/__init__.py +12 -0
  638. executorch/backends/cortex_m/passes/activation_fusion_pass.py +181 -0
  639. executorch/backends/cortex_m/passes/clamp_hardswish_pass.py +37 -0
  640. executorch/backends/cortex_m/passes/convert_to_cortex_m_pass.py +318 -0
  641. executorch/backends/cortex_m/passes/cortex_m_pass_manager.py +73 -0
  642. executorch/backends/cortex_m/passes/decompose_hardswish_pass.py +127 -0
  643. executorch/backends/cortex_m/passes/passes_utils.py +226 -0
  644. executorch/backends/cortex_m/passes/quantized_op_fusion_pass.py +255 -0
  645. executorch/backends/cortex_m/passes/replace_quant_nodes_pass.py +67 -0
  646. executorch/backends/cortex_m/quantizer/__init__.py +18 -0
  647. executorch/backends/cortex_m/quantizer/operator_configs.py +74 -0
  648. executorch/backends/cortex_m/quantizer/quantization_configs.py +121 -0
  649. executorch/backends/cortex_m/quantizer/quantizer.py +500 -0
  650. executorch/backends/cortex_m/test/misc/test_quantization.py +359 -0
  651. executorch/backends/cortex_m/test/models/__init__.py +4 -0
  652. executorch/backends/cortex_m/test/models/test_mobilenet_v3.py +74 -0
  653. executorch/backends/cortex_m/test/ops/__init__.py +4 -0
  654. executorch/backends/cortex_m/test/ops/test_activation.py +528 -0
  655. executorch/backends/cortex_m/test/ops/test_add.py +182 -0
  656. executorch/backends/cortex_m/test/ops/test_avg_pool2d.py +99 -0
  657. executorch/backends/cortex_m/test/ops/test_conv.py +342 -0
  658. executorch/backends/cortex_m/test/ops/test_linear.py +130 -0
  659. executorch/backends/cortex_m/test/ops/test_lstm.py +98 -0
  660. executorch/backends/cortex_m/test/ops/test_maximum.py +83 -0
  661. executorch/backends/cortex_m/test/ops/test_minimum.py +104 -0
  662. executorch/backends/cortex_m/test/ops/test_mul.py +156 -0
  663. executorch/backends/cortex_m/test/ops/test_softmax.py +84 -0
  664. executorch/backends/cortex_m/test/ops/test_transpose.py +102 -0
  665. executorch/backends/cortex_m/test/test_helpers_passes_utils.py +103 -0
  666. executorch/backends/cortex_m/test/test_replace_quant_nodes.py +134 -0
  667. executorch/backends/cortex_m/test/tester.py +113 -0
  668. executorch/backends/cuda/__init__.py +5 -0
  669. executorch/backends/cuda/cuda_backend.py +250 -0
  670. executorch/backends/cuda/cuda_partitioner.py +25 -0
  671. executorch/backends/cuda/passes/__init__.py +0 -0
  672. executorch/backends/cuda/passes/move_cond_predicate_to_cpu.py +90 -0
  673. executorch/backends/cuda/passes/tests/__init__.py +0 -0
  674. executorch/backends/cuda/passes/tests/test_move_cond_predicate_to_cpu.py +553 -0
  675. executorch/backends/cuda/tests/__init__.py +5 -0
  676. executorch/backends/cuda/tests/test_cuda_export.py +327 -0
  677. executorch/backends/cuda/tests/test_cuda_partitioner.py +224 -0
  678. executorch/backends/cuda/triton/__init__.py +17 -0
  679. executorch/backends/cuda/triton/kernels/__init__.py +11 -0
  680. executorch/backends/cuda/triton/kernels/sdpa.py +827 -0
  681. executorch/backends/cuda/triton/replacement_pass.py +130 -0
  682. executorch/backends/example/example_backend.py +41 -0
  683. executorch/backends/example/example_backend_delegate_passes/merge_to_dim_pass.py +83 -0
  684. executorch/backends/example/example_backend_delegate_passes/permute_memory_formats_pass.py +143 -0
  685. executorch/backends/example/example_operators/adaptive_avg_pool2d.py +47 -0
  686. executorch/backends/example/example_operators/add.py +44 -0
  687. executorch/backends/example/example_operators/conv2d.py +52 -0
  688. executorch/backends/example/example_operators/conv_relu.py +49 -0
  689. executorch/backends/example/example_operators/dropout.py +44 -0
  690. executorch/backends/example/example_operators/flatten.py +40 -0
  691. executorch/backends/example/example_operators/linear.py +62 -0
  692. executorch/backends/example/example_operators/op_base.py +15 -0
  693. executorch/backends/example/example_operators/ops.py +32 -0
  694. executorch/backends/example/example_operators/utils.py +34 -0
  695. executorch/backends/example/example_partitioner.py +88 -0
  696. executorch/backends/example/example_quantizer.py +108 -0
  697. executorch/backends/example/test_example_delegate.py +112 -0
  698. executorch/backends/mediatek/__init__.py +5 -0
  699. executorch/backends/mediatek/_passes/__init__.py +0 -0
  700. executorch/backends/mediatek/_passes/decompose_scaled_dot_product_attention.py +79 -0
  701. executorch/backends/mediatek/partitioner.py +107 -0
  702. executorch/backends/mediatek/preprocess.py +214 -0
  703. executorch/backends/mediatek/quantizer/__init__.py +4 -0
  704. executorch/backends/mediatek/quantizer/annotator.py +263 -0
  705. executorch/backends/mediatek/quantizer/qconfig.py +174 -0
  706. executorch/backends/mediatek/quantizer/quantizer.py +45 -0
  707. executorch/backends/nxp/_passes/remove_getitem_pass.py +105 -0
  708. executorch/backends/nxp/aten_passes/convert_unsqueeze_to_view.py +79 -0
  709. executorch/backends/nxp/aten_passes/decompose_split_to_slices_pass.py +193 -0
  710. executorch/backends/nxp/aten_passes/fuse_batch_norm_with_conv_pass.py +146 -0
  711. executorch/backends/nxp/aten_passes/fuse_batch_norm_with_linear_pass.py +150 -0
  712. executorch/backends/nxp/aten_passes/fuse_linear_and_add_pass.py +204 -0
  713. executorch/backends/nxp/aten_passes/move_activation_before_concat.py +102 -0
  714. executorch/backends/nxp/aten_passes/neutron_aten_pass_manager.py +80 -0
  715. executorch/backends/nxp/aten_passes/remove_nodes_with_known_outputs.py +196 -0
  716. executorch/backends/nxp/aten_passes/split_group_convolution.py +281 -0
  717. executorch/backends/nxp/aten_passes/split_gru_based_on_num_layers.py +307 -0
  718. executorch/backends/nxp/backend/custom_delegation_options.py +19 -0
  719. executorch/backends/nxp/backend/edge_helper.py +222 -0
  720. executorch/backends/nxp/backend/edge_program_converter.py +247 -0
  721. executorch/backends/nxp/backend/ir/conversion_config.py +54 -0
  722. executorch/backends/nxp/backend/ir/conversion_context.py +38 -0
  723. executorch/backends/nxp/backend/ir/converter/__init__.py +0 -0
  724. executorch/backends/nxp/backend/ir/converter/builder/__init__.py +0 -0
  725. executorch/backends/nxp/backend/ir/converter/builder/aten_model_builder_director.py +142 -0
  726. executorch/backends/nxp/backend/ir/converter/builder/model_builder.py +1720 -0
  727. executorch/backends/nxp/backend/ir/converter/builder/quantization_verification.py +377 -0
  728. executorch/backends/nxp/backend/ir/converter/conversion/__init__.py +0 -0
  729. executorch/backends/nxp/backend/ir/converter/conversion/aten_translator.py +71 -0
  730. executorch/backends/nxp/backend/ir/converter/conversion/common.py +207 -0
  731. executorch/backends/nxp/backend/ir/converter/conversion/translator.py +689 -0
  732. executorch/backends/nxp/backend/ir/converter/node_converter.py +234 -0
  733. executorch/backends/nxp/backend/ir/converter/node_converters/__init__.py +4 -0
  734. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/__init__.py +101 -0
  735. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/abs_converter.py +35 -0
  736. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/adaptive_avg_pool_2d_converter.py +68 -0
  737. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/add_tensor_converter.py +57 -0
  738. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/addmm_converter.py +62 -0
  739. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/avg_pool_2d_converter.py +101 -0
  740. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/cat_converter.py +209 -0
  741. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/clone_converter.py +118 -0
  742. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/constant_pad_nd_converter.py +160 -0
  743. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/convolution_converter.py +535 -0
  744. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/hardtanh_converter.py +47 -0
  745. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/max_pool_2d_converter.py +104 -0
  746. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mean_dim_converter.py +112 -0
  747. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mm_converter.py +59 -0
  748. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mul_tensor_converter.py +61 -0
  749. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/permute_copy_converter.py +468 -0
  750. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/qdq_dequantize_converter.py +116 -0
  751. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/qdq_quantize_converter.py +46 -0
  752. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/relu_converter.py +33 -0
  753. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/sigmoid_converter.py +35 -0
  754. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/slice_tensor_converter.py +158 -0
  755. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/softmax_converter.py +62 -0
  756. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/sub_tensor_converter.py +59 -0
  757. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/tanh_converter.py +33 -0
  758. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/view_copy_converter.py +220 -0
  759. executorch/backends/nxp/backend/ir/converter/node_converters/shared/__init__.py +0 -0
  760. executorch/backends/nxp/backend/ir/converter/node_converters/shared/conv_utils.py +105 -0
  761. executorch/backends/nxp/backend/ir/converter/node_converters/shared/recurrent_utils.py +64 -0
  762. executorch/backends/nxp/backend/ir/converter/node_converters/shared/reduce_utils.py +97 -0
  763. executorch/backends/nxp/backend/ir/converter/node_converters/shared/reshape_transposition.py +233 -0
  764. executorch/backends/nxp/backend/ir/converter/quantization_utils.py +149 -0
  765. executorch/backends/nxp/backend/ir/converter/tensor_utils.py +50 -0
  766. executorch/backends/nxp/backend/ir/lib/__init__.py +0 -0
  767. executorch/backends/nxp/backend/ir/lib/tflite/ATan2Options.py +50 -0
  768. executorch/backends/nxp/backend/ir/lib/tflite/AbsOptions.py +50 -0
  769. executorch/backends/nxp/backend/ir/lib/tflite/ActivationFunctionType.py +12 -0
  770. executorch/backends/nxp/backend/ir/lib/tflite/AddNOptions.py +50 -0
  771. executorch/backends/nxp/backend/ir/lib/tflite/AddOptions.py +82 -0
  772. executorch/backends/nxp/backend/ir/lib/tflite/ArgMaxOptions.py +65 -0
  773. executorch/backends/nxp/backend/ir/lib/tflite/ArgMinOptions.py +65 -0
  774. executorch/backends/nxp/backend/ir/lib/tflite/AssignVariableOptions.py +50 -0
  775. executorch/backends/nxp/backend/ir/lib/tflite/BatchMatMulOptions.py +101 -0
  776. executorch/backends/nxp/backend/ir/lib/tflite/BatchToSpaceNDOptions.py +50 -0
  777. executorch/backends/nxp/backend/ir/lib/tflite/BidirectionalSequenceLSTMOptions.py +160 -0
  778. executorch/backends/nxp/backend/ir/lib/tflite/BidirectionalSequenceRNNOptions.py +126 -0
  779. executorch/backends/nxp/backend/ir/lib/tflite/BitcastOptions.py +50 -0
  780. executorch/backends/nxp/backend/ir/lib/tflite/BitwiseXorOptions.py +50 -0
  781. executorch/backends/nxp/backend/ir/lib/tflite/BroadcastToOptions.py +50 -0
  782. executorch/backends/nxp/backend/ir/lib/tflite/BucketizeOptions.py +98 -0
  783. executorch/backends/nxp/backend/ir/lib/tflite/Buffer.py +132 -0
  784. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOperator.py +212 -0
  785. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOptions.py +133 -0
  786. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOptions2.py +27 -0
  787. executorch/backends/nxp/backend/ir/lib/tflite/CallOnceOptions.py +65 -0
  788. executorch/backends/nxp/backend/ir/lib/tflite/CallOptions.py +67 -0
  789. executorch/backends/nxp/backend/ir/lib/tflite/CastOptions.py +80 -0
  790. executorch/backends/nxp/backend/ir/lib/tflite/CombinerType.py +9 -0
  791. executorch/backends/nxp/backend/ir/lib/tflite/ConcatEmbeddingsOptions.py +163 -0
  792. executorch/backends/nxp/backend/ir/lib/tflite/ConcatenationOptions.py +80 -0
  793. executorch/backends/nxp/backend/ir/lib/tflite/Conv2DOptions.py +155 -0
  794. executorch/backends/nxp/backend/ir/lib/tflite/Conv3DOptions.py +170 -0
  795. executorch/backends/nxp/backend/ir/lib/tflite/CosOptions.py +50 -0
  796. executorch/backends/nxp/backend/ir/lib/tflite/CumsumOptions.py +84 -0
  797. executorch/backends/nxp/backend/ir/lib/tflite/CustomOptionsFormat.py +7 -0
  798. executorch/backends/nxp/backend/ir/lib/tflite/CustomQuantization.py +98 -0
  799. executorch/backends/nxp/backend/ir/lib/tflite/DensifyOptions.py +50 -0
  800. executorch/backends/nxp/backend/ir/lib/tflite/DepthToSpaceOptions.py +65 -0
  801. executorch/backends/nxp/backend/ir/lib/tflite/DepthwiseConv2DOptions.py +157 -0
  802. executorch/backends/nxp/backend/ir/lib/tflite/DequantizeOptions.py +50 -0
  803. executorch/backends/nxp/backend/ir/lib/tflite/DilateOptions.py +50 -0
  804. executorch/backends/nxp/backend/ir/lib/tflite/DimensionMetadata.py +152 -0
  805. executorch/backends/nxp/backend/ir/lib/tflite/DimensionType.py +8 -0
  806. executorch/backends/nxp/backend/ir/lib/tflite/DivOptions.py +65 -0
  807. executorch/backends/nxp/backend/ir/lib/tflite/DynamicUpdateSliceOptions.py +52 -0
  808. executorch/backends/nxp/backend/ir/lib/tflite/EmbeddingLookupSparseOptions.py +67 -0
  809. executorch/backends/nxp/backend/ir/lib/tflite/EqualOptions.py +50 -0
  810. executorch/backends/nxp/backend/ir/lib/tflite/ExpOptions.py +50 -0
  811. executorch/backends/nxp/backend/ir/lib/tflite/ExpandDimsOptions.py +50 -0
  812. executorch/backends/nxp/backend/ir/lib/tflite/FakeQuantOptions.py +116 -0
  813. executorch/backends/nxp/backend/ir/lib/tflite/FillOptions.py +50 -0
  814. executorch/backends/nxp/backend/ir/lib/tflite/FloorDivOptions.py +50 -0
  815. executorch/backends/nxp/backend/ir/lib/tflite/FloorModOptions.py +50 -0
  816. executorch/backends/nxp/backend/ir/lib/tflite/FullyConnectedOptions.py +129 -0
  817. executorch/backends/nxp/backend/ir/lib/tflite/FullyConnectedOptionsWeightsFormat.py +8 -0
  818. executorch/backends/nxp/backend/ir/lib/tflite/GatherNdOptions.py +50 -0
  819. executorch/backends/nxp/backend/ir/lib/tflite/GatherOptions.py +80 -0
  820. executorch/backends/nxp/backend/ir/lib/tflite/GeluOptions.py +67 -0
  821. executorch/backends/nxp/backend/ir/lib/tflite/GreaterEqualOptions.py +50 -0
  822. executorch/backends/nxp/backend/ir/lib/tflite/GreaterOptions.py +50 -0
  823. executorch/backends/nxp/backend/ir/lib/tflite/HardSwishOptions.py +50 -0
  824. executorch/backends/nxp/backend/ir/lib/tflite/HashtableFindOptions.py +50 -0
  825. executorch/backends/nxp/backend/ir/lib/tflite/HashtableImportOptions.py +52 -0
  826. executorch/backends/nxp/backend/ir/lib/tflite/HashtableOptions.py +95 -0
  827. executorch/backends/nxp/backend/ir/lib/tflite/HashtableSizeOptions.py +50 -0
  828. executorch/backends/nxp/backend/ir/lib/tflite/IfOptions.py +80 -0
  829. executorch/backends/nxp/backend/ir/lib/tflite/Int32Vector.py +98 -0
  830. executorch/backends/nxp/backend/ir/lib/tflite/L2NormOptions.py +65 -0
  831. executorch/backends/nxp/backend/ir/lib/tflite/LSHProjectionOptions.py +65 -0
  832. executorch/backends/nxp/backend/ir/lib/tflite/LSHProjectionType.py +9 -0
  833. executorch/backends/nxp/backend/ir/lib/tflite/LSTMKernelType.py +8 -0
  834. executorch/backends/nxp/backend/ir/lib/tflite/LSTMOptions.py +131 -0
  835. executorch/backends/nxp/backend/ir/lib/tflite/LeakyReluOptions.py +67 -0
  836. executorch/backends/nxp/backend/ir/lib/tflite/LessEqualOptions.py +50 -0
  837. executorch/backends/nxp/backend/ir/lib/tflite/LessOptions.py +50 -0
  838. executorch/backends/nxp/backend/ir/lib/tflite/LocalResponseNormalizationOptions.py +118 -0
  839. executorch/backends/nxp/backend/ir/lib/tflite/LogSoftmaxOptions.py +50 -0
  840. executorch/backends/nxp/backend/ir/lib/tflite/LogicalAndOptions.py +50 -0
  841. executorch/backends/nxp/backend/ir/lib/tflite/LogicalNotOptions.py +50 -0
  842. executorch/backends/nxp/backend/ir/lib/tflite/LogicalOrOptions.py +50 -0
  843. executorch/backends/nxp/backend/ir/lib/tflite/MatrixDiagOptions.py +50 -0
  844. executorch/backends/nxp/backend/ir/lib/tflite/MatrixSetDiagOptions.py +50 -0
  845. executorch/backends/nxp/backend/ir/lib/tflite/MaximumMinimumOptions.py +50 -0
  846. executorch/backends/nxp/backend/ir/lib/tflite/Metadata.py +84 -0
  847. executorch/backends/nxp/backend/ir/lib/tflite/MirrorPadMode.py +8 -0
  848. executorch/backends/nxp/backend/ir/lib/tflite/MirrorPadOptions.py +65 -0
  849. executorch/backends/nxp/backend/ir/lib/tflite/Model.py +352 -0
  850. executorch/backends/nxp/backend/ir/lib/tflite/MulOptions.py +65 -0
  851. executorch/backends/nxp/backend/ir/lib/tflite/NegOptions.py +50 -0
  852. executorch/backends/nxp/backend/ir/lib/tflite/NonMaxSuppressionV4Options.py +52 -0
  853. executorch/backends/nxp/backend/ir/lib/tflite/NonMaxSuppressionV5Options.py +52 -0
  854. executorch/backends/nxp/backend/ir/lib/tflite/NotEqualOptions.py +50 -0
  855. executorch/backends/nxp/backend/ir/lib/tflite/OneHotOptions.py +65 -0
  856. executorch/backends/nxp/backend/ir/lib/tflite/Operator.py +428 -0
  857. executorch/backends/nxp/backend/ir/lib/tflite/OperatorCode.py +112 -0
  858. executorch/backends/nxp/backend/ir/lib/tflite/PackOptions.py +80 -0
  859. executorch/backends/nxp/backend/ir/lib/tflite/PadOptions.py +50 -0
  860. executorch/backends/nxp/backend/ir/lib/tflite/PadV2Options.py +50 -0
  861. executorch/backends/nxp/backend/ir/lib/tflite/Padding.py +8 -0
  862. executorch/backends/nxp/backend/ir/lib/tflite/Pool2DOptions.py +140 -0
  863. executorch/backends/nxp/backend/ir/lib/tflite/PowOptions.py +50 -0
  864. executorch/backends/nxp/backend/ir/lib/tflite/QuantizationDetails.py +8 -0
  865. executorch/backends/nxp/backend/ir/lib/tflite/QuantizationParameters.py +295 -0
  866. executorch/backends/nxp/backend/ir/lib/tflite/QuantizeOptions.py +50 -0
  867. executorch/backends/nxp/backend/ir/lib/tflite/RNNOptions.py +82 -0
  868. executorch/backends/nxp/backend/ir/lib/tflite/RandomOptions.py +80 -0
  869. executorch/backends/nxp/backend/ir/lib/tflite/RangeOptions.py +50 -0
  870. executorch/backends/nxp/backend/ir/lib/tflite/RankOptions.py +50 -0
  871. executorch/backends/nxp/backend/ir/lib/tflite/ReadVariableOptions.py +50 -0
  872. executorch/backends/nxp/backend/ir/lib/tflite/ReduceWindowFunction.py +13 -0
  873. executorch/backends/nxp/backend/ir/lib/tflite/ReduceWindowOptions.py +65 -0
  874. executorch/backends/nxp/backend/ir/lib/tflite/ReducerOptions.py +67 -0
  875. executorch/backends/nxp/backend/ir/lib/tflite/ReshapeOptions.py +98 -0
  876. executorch/backends/nxp/backend/ir/lib/tflite/ResizeBilinearOptions.py +84 -0
  877. executorch/backends/nxp/backend/ir/lib/tflite/ResizeNearestNeighborOptions.py +86 -0
  878. executorch/backends/nxp/backend/ir/lib/tflite/ReverseSequenceOptions.py +82 -0
  879. executorch/backends/nxp/backend/ir/lib/tflite/ReverseV2Options.py +50 -0
  880. executorch/backends/nxp/backend/ir/lib/tflite/Rfft2dOptions.py +50 -0
  881. executorch/backends/nxp/backend/ir/lib/tflite/RightShiftOptions.py +50 -0
  882. executorch/backends/nxp/backend/ir/lib/tflite/RngAlgorithm.py +9 -0
  883. executorch/backends/nxp/backend/ir/lib/tflite/SVDFOptions.py +97 -0
  884. executorch/backends/nxp/backend/ir/lib/tflite/ScatterNdOptions.py +50 -0
  885. executorch/backends/nxp/backend/ir/lib/tflite/SegmentSumOptions.py +50 -0
  886. executorch/backends/nxp/backend/ir/lib/tflite/SelectOptions.py +50 -0
  887. executorch/backends/nxp/backend/ir/lib/tflite/SelectV2Options.py +50 -0
  888. executorch/backends/nxp/backend/ir/lib/tflite/SequenceRNNOptions.py +99 -0
  889. executorch/backends/nxp/backend/ir/lib/tflite/ShapeOptions.py +65 -0
  890. executorch/backends/nxp/backend/ir/lib/tflite/SignOptions.py +50 -0
  891. executorch/backends/nxp/backend/ir/lib/tflite/SignatureDef.py +172 -0
  892. executorch/backends/nxp/backend/ir/lib/tflite/SkipGramOptions.py +97 -0
  893. executorch/backends/nxp/backend/ir/lib/tflite/SliceOptions.py +50 -0
  894. executorch/backends/nxp/backend/ir/lib/tflite/SoftmaxOptions.py +67 -0
  895. executorch/backends/nxp/backend/ir/lib/tflite/SpaceToBatchNDOptions.py +50 -0
  896. executorch/backends/nxp/backend/ir/lib/tflite/SpaceToDepthOptions.py +65 -0
  897. executorch/backends/nxp/backend/ir/lib/tflite/SparseIndexVector.py +10 -0
  898. executorch/backends/nxp/backend/ir/lib/tflite/SparseToDenseOptions.py +67 -0
  899. executorch/backends/nxp/backend/ir/lib/tflite/SparsityParameters.py +190 -0
  900. executorch/backends/nxp/backend/ir/lib/tflite/SplitOptions.py +65 -0
  901. executorch/backends/nxp/backend/ir/lib/tflite/SplitVOptions.py +65 -0
  902. executorch/backends/nxp/backend/ir/lib/tflite/SquareOptions.py +50 -0
  903. executorch/backends/nxp/backend/ir/lib/tflite/SquaredDifferenceOptions.py +52 -0
  904. executorch/backends/nxp/backend/ir/lib/tflite/SqueezeOptions.py +98 -0
  905. executorch/backends/nxp/backend/ir/lib/tflite/StablehloBroadcastInDimOptions.py +102 -0
  906. executorch/backends/nxp/backend/ir/lib/tflite/StablehloCompareOptions.py +86 -0
  907. executorch/backends/nxp/backend/ir/lib/tflite/StablehloComparisonDirection.py +12 -0
  908. executorch/backends/nxp/backend/ir/lib/tflite/StablehloComparisonType.py +11 -0
  909. executorch/backends/nxp/backend/ir/lib/tflite/StablehloConcatenateOptions.py +67 -0
  910. executorch/backends/nxp/backend/ir/lib/tflite/StablehloConvolutionOptions.py +634 -0
  911. executorch/backends/nxp/backend/ir/lib/tflite/StablehloCustomCallOptions.py +214 -0
  912. executorch/backends/nxp/backend/ir/lib/tflite/StablehloDotGeneralOptions.py +304 -0
  913. executorch/backends/nxp/backend/ir/lib/tflite/StablehloDynamicSliceOptions.py +100 -0
  914. executorch/backends/nxp/backend/ir/lib/tflite/StablehloGatherOptions.py +276 -0
  915. executorch/backends/nxp/backend/ir/lib/tflite/StablehloIotaOptions.py +65 -0
  916. executorch/backends/nxp/backend/ir/lib/tflite/StablehloPadOptions.py +194 -0
  917. executorch/backends/nxp/backend/ir/lib/tflite/StablehloPrecisionConfig.py +9 -0
  918. executorch/backends/nxp/backend/ir/lib/tflite/StablehloReduceOptions.py +115 -0
  919. executorch/backends/nxp/backend/ir/lib/tflite/StablehloReduceWindowOptions.py +307 -0
  920. executorch/backends/nxp/backend/ir/lib/tflite/StablehloRngBitGeneratorOptions.py +67 -0
  921. executorch/backends/nxp/backend/ir/lib/tflite/StablehloScatterOptions.py +268 -0
  922. executorch/backends/nxp/backend/ir/lib/tflite/StablehloSliceOptions.py +194 -0
  923. executorch/backends/nxp/backend/ir/lib/tflite/StablehloSortOptions.py +97 -0
  924. executorch/backends/nxp/backend/ir/lib/tflite/StablehloTransposeOptions.py +100 -0
  925. executorch/backends/nxp/backend/ir/lib/tflite/StablehloWhileOptions.py +80 -0
  926. executorch/backends/nxp/backend/ir/lib/tflite/StridedSliceOptions.py +142 -0
  927. executorch/backends/nxp/backend/ir/lib/tflite/SubGraph.py +251 -0
  928. executorch/backends/nxp/backend/ir/lib/tflite/SubOptions.py +82 -0
  929. executorch/backends/nxp/backend/ir/lib/tflite/Tensor.py +317 -0
  930. executorch/backends/nxp/backend/ir/lib/tflite/TensorMap.py +84 -0
  931. executorch/backends/nxp/backend/ir/lib/tflite/TensorType.py +24 -0
  932. executorch/backends/nxp/backend/ir/lib/tflite/TileOptions.py +50 -0
  933. executorch/backends/nxp/backend/ir/lib/tflite/TopKV2Options.py +50 -0
  934. executorch/backends/nxp/backend/ir/lib/tflite/TransposeConvOptions.py +125 -0
  935. executorch/backends/nxp/backend/ir/lib/tflite/TransposeOptions.py +50 -0
  936. executorch/backends/nxp/backend/ir/lib/tflite/Uint16Vector.py +98 -0
  937. executorch/backends/nxp/backend/ir/lib/tflite/Uint8Vector.py +98 -0
  938. executorch/backends/nxp/backend/ir/lib/tflite/UnidirectionalSequenceLSTMOptions.py +164 -0
  939. executorch/backends/nxp/backend/ir/lib/tflite/UniqueOptions.py +65 -0
  940. executorch/backends/nxp/backend/ir/lib/tflite/UnpackOptions.py +80 -0
  941. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentMaxOptions.py +52 -0
  942. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentMinOptions.py +52 -0
  943. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentProdOptions.py +52 -0
  944. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentSumOptions.py +52 -0
  945. executorch/backends/nxp/backend/ir/lib/tflite/VarHandleOptions.py +84 -0
  946. executorch/backends/nxp/backend/ir/lib/tflite/VariantSubType.py +130 -0
  947. executorch/backends/nxp/backend/ir/lib/tflite/WhereOptions.py +50 -0
  948. executorch/backends/nxp/backend/ir/lib/tflite/WhileOptions.py +80 -0
  949. executorch/backends/nxp/backend/ir/lib/tflite/ZerosLikeOptions.py +50 -0
  950. executorch/backends/nxp/backend/ir/lib/tflite/__init__.py +0 -0
  951. executorch/backends/nxp/backend/ir/logger.py +341 -0
  952. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/__init__.py +0 -0
  953. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/graph_utils.py +115 -0
  954. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/operator_rules.py +102 -0
  955. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/__init__.py +0 -0
  956. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/base_optimization.py +41 -0
  957. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/permute_fully_connected_weights_after_reshape.py +121 -0
  958. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/prune_transpose_operators.py +174 -0
  959. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizer.py +127 -0
  960. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/pattern_matcher.py +925 -0
  961. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/tensor_rules.py +710 -0
  962. executorch/backends/nxp/backend/ir/tensor_formatting.py +69 -0
  963. executorch/backends/nxp/backend/ir/tflite_generator/__init__.py +0 -0
  964. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/__init__.py +0 -0
  965. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/abs_options.py +25 -0
  966. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/add_n_options.py +32 -0
  967. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/add_options.py +46 -0
  968. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/arg_max_options.py +28 -0
  969. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/arg_min_options.py +28 -0
  970. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/average_pool_2d_options.py +63 -0
  971. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/batch_mat_mul_options.py +45 -0
  972. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bidirectional_sequence_lstm_options.py +65 -0
  973. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bidirectional_sequence_rnn_options.py +53 -0
  974. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bitwise_xor_options.py +23 -0
  975. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/broadcast_to_options.py +25 -0
  976. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/cast_options.py +32 -0
  977. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/concatenation_options.py +38 -0
  978. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/conv_2d_options.py +59 -0
  979. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/conv_3d_options.py +67 -0
  980. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/cum_sum_options.py +30 -0
  981. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/depth_to_space_options.py +29 -0
  982. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/depthwise_conv_2d_options.py +69 -0
  983. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/dequantize_options.py +24 -0
  984. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/div_options.py +43 -0
  985. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/equal_options.py +23 -0
  986. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/exp_options.py +23 -0
  987. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/floor_mod_options.py +23 -0
  988. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/fully_connected_options.py +57 -0
  989. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gather_nd_options.py +23 -0
  990. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gather_options.py +30 -0
  991. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gelu_options.py +27 -0
  992. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/greater_equal_options.py +25 -0
  993. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/greater_options.py +23 -0
  994. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/hard_swish_options.py +23 -0
  995. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/leaky_relu_options.py +32 -0
  996. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/less_equal_options.py +23 -0
  997. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/less_options.py +23 -0
  998. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/log_softmax_options.py +30 -0
  999. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_and_options.py +23 -0
  1000. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_not_options.py +23 -0
  1001. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_or_options.py +23 -0
  1002. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/lrn_options.py +45 -0
  1003. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/lstm_options.py +60 -0
  1004. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/max_pool_2d_options.py +63 -0
  1005. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/maximum_options.py +21 -0
  1006. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mean_options.py +28 -0
  1007. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/minimum_options.py +21 -0
  1008. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mirror_pad_options.py +60 -0
  1009. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mul_options.py +36 -0
  1010. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/multinomial_options.py +30 -0
  1011. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/neg_options.py +23 -0
  1012. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/not_equal_options.py +23 -0
  1013. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/one_hot_options.py +27 -0
  1014. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pad_options.py +27 -0
  1015. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pad_v2_options.py +27 -0
  1016. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pow_options.py +24 -0
  1017. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/quantize_options.py +30 -0
  1018. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/range_options.py +23 -0
  1019. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_max_options.py +28 -0
  1020. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_min_options.py +28 -0
  1021. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_prod_options.py +28 -0
  1022. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reshape_options.py +53 -0
  1023. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/resize_bilinear_options.py +34 -0
  1024. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/resize_nearest_neighbor_options.py +37 -0
  1025. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reverse_sequence_options.py +32 -0
  1026. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/scatter_nd_options.py +23 -0
  1027. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/select_v2_options.py +24 -0
  1028. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/shape_options.py +34 -0
  1029. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sign_options.py +22 -0
  1030. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/slice_options.py +29 -0
  1031. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/softmax_options.py +36 -0
  1032. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/space_to_depth_options.py +29 -0
  1033. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/split_options.py +27 -0
  1034. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/split_v_options.py +27 -0
  1035. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/square_options.py +23 -0
  1036. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/squared_difference_options.py +25 -0
  1037. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/squeeze_options.py +42 -0
  1038. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/strided_slice_options.py +35 -0
  1039. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sub_options.py +46 -0
  1040. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sum_options.py +28 -0
  1041. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/tile_options.py +23 -0
  1042. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/transpose_conv_options.py +50 -0
  1043. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/transpose_options.py +30 -0
  1044. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/unidirectional_sequence_lstm_options.py +67 -0
  1045. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/unidirectional_sequence_rnn_options.py +49 -0
  1046. executorch/backends/nxp/backend/ir/tflite_generator/custom_options/flex_transpose_options.py +78 -0
  1047. executorch/backends/nxp/backend/ir/tflite_generator/meta/__init__.py +0 -0
  1048. executorch/backends/nxp/backend/ir/tflite_generator/meta/meta.py +255 -0
  1049. executorch/backends/nxp/backend/ir/tflite_generator/meta/types.py +198 -0
  1050. executorch/backends/nxp/backend/ir/tflite_generator/tflite_model.py +814 -0
  1051. executorch/backends/nxp/backend/neutron_converter_manager.py +117 -0
  1052. executorch/backends/nxp/backend/neutron_operator_support.py +79 -0
  1053. executorch/backends/nxp/backend/neutron_target_spec.py +143 -0
  1054. executorch/backends/nxp/backend/node_format.py +26 -0
  1055. executorch/backends/nxp/backend/node_format_inference.py +294 -0
  1056. executorch/backends/nxp/edge_passes/move_auxiliary_operator_into_separate_qdq_cluster_pass.py +254 -0
  1057. executorch/backends/nxp/edge_passes/neutron_edge_pass.py +55 -0
  1058. executorch/backends/nxp/edge_passes/neutron_edge_pass_manager.py +29 -0
  1059. executorch/backends/nxp/edge_passes/remove_additional_quantize_dequantize_nodes_pass.py +111 -0
  1060. executorch/backends/nxp/edge_passes/remove_as_strided_copy_nodes.py +66 -0
  1061. executorch/backends/nxp/edge_passes/remove_io_quant_ops_pass.py +79 -0
  1062. executorch/backends/nxp/neutron_node_extraction.py +152 -0
  1063. executorch/backends/nxp/neutron_partitioner.py +421 -0
  1064. executorch/backends/nxp/neutron_pass_manager.py +50 -0
  1065. executorch/backends/nxp/nxp_backend.py +391 -0
  1066. executorch/backends/nxp/quantizer/neutron_quantizer.py +353 -0
  1067. executorch/backends/nxp/quantizer/patterns.py +1089 -0
  1068. executorch/backends/nxp/quantizer/utils.py +195 -0
  1069. executorch/backends/nxp/tests/executorch_pipeline.py +184 -0
  1070. executorch/backends/nxp/tests/executors.py +401 -0
  1071. executorch/backends/nxp/tests/ir/__init__.py +0 -0
  1072. executorch/backends/nxp/tests/ir/converter/__init__.py +0 -0
  1073. executorch/backends/nxp/tests/ir/converter/node_converter/__init__.py +0 -0
  1074. executorch/backends/nxp/tests/ir/converter/node_converter/test_abs_converter.py +91 -0
  1075. executorch/backends/nxp/tests/ir/converter/node_converter/test_adaptive_avg_pool2d_converter.py +148 -0
  1076. executorch/backends/nxp/tests/ir/converter/node_converter/test_add_tensor_converter.py +161 -0
  1077. executorch/backends/nxp/tests/ir/converter/node_converter/test_addmm_converter.py +96 -0
  1078. executorch/backends/nxp/tests/ir/converter/node_converter/test_avg_pool2d_converter.py +221 -0
  1079. executorch/backends/nxp/tests/ir/converter/node_converter/test_cat_converter.py +495 -0
  1080. executorch/backends/nxp/tests/ir/converter/node_converter/test_clone_converter.py +344 -0
  1081. executorch/backends/nxp/tests/ir/converter/node_converter/test_constant_pad_nd_converter.py +188 -0
  1082. executorch/backends/nxp/tests/ir/converter/node_converter/test_conv_converter.py +680 -0
  1083. executorch/backends/nxp/tests/ir/converter/node_converter/test_hardtanh_converter.py +107 -0
  1084. executorch/backends/nxp/tests/ir/converter/node_converter/test_max_pool_2d_converter.py +132 -0
  1085. executorch/backends/nxp/tests/ir/converter/node_converter/test_mean_dim_converter.py +280 -0
  1086. executorch/backends/nxp/tests/ir/converter/node_converter/test_mm_converter.py +96 -0
  1087. executorch/backends/nxp/tests/ir/converter/node_converter/test_mul_tensor_converter.py +212 -0
  1088. executorch/backends/nxp/tests/ir/converter/node_converter/test_permute_copy_converter.py +426 -0
  1089. executorch/backends/nxp/tests/ir/converter/node_converter/test_relu_converter.py +114 -0
  1090. executorch/backends/nxp/tests/ir/converter/node_converter/test_sigmoid_converter.py +78 -0
  1091. executorch/backends/nxp/tests/ir/converter/node_converter/test_slice_tensor_converter.py +244 -0
  1092. executorch/backends/nxp/tests/ir/converter/node_converter/test_softmax_converter.py +115 -0
  1093. executorch/backends/nxp/tests/ir/converter/node_converter/test_sub_tensor_converter.py +181 -0
  1094. executorch/backends/nxp/tests/ir/converter/node_converter/test_tanh_converter.py +90 -0
  1095. executorch/backends/nxp/tests/ir/converter/node_converter/test_view_copy_converter.py +509 -0
  1096. executorch/backends/nxp/tests/ir/edge_passes/__init__.py +0 -0
  1097. executorch/backends/nxp/tests/ir/edge_passes/test_remove_io_quant_ops_pass.py +127 -0
  1098. executorch/backends/nxp/tests/models.py +737 -0
  1099. executorch/backends/nxp/tests/test_batch_norm_fusion.py +233 -0
  1100. executorch/backends/nxp/tests/test_context_sensitive_delegation.py +71 -0
  1101. executorch/backends/nxp/tests/test_convert_unsqueeze_to_view.py +143 -0
  1102. executorch/backends/nxp/tests/test_decompose_split_to_slices.py +315 -0
  1103. executorch/backends/nxp/tests/test_edge_passes.py +382 -0
  1104. executorch/backends/nxp/tests/test_gru_splitting.py +267 -0
  1105. executorch/backends/nxp/tests/test_integration.py +53 -0
  1106. executorch/backends/nxp/tests/test_linear_and_add_fusion.py +653 -0
  1107. executorch/backends/nxp/tests/test_move_activation_before_concatenation.py +905 -0
  1108. executorch/backends/nxp/tests/test_neutron_backend.py +63 -0
  1109. executorch/backends/nxp/tests/test_neutron_backend_executor.py +327 -0
  1110. executorch/backends/nxp/tests/test_neutron_converter_manager.py +59 -0
  1111. executorch/backends/nxp/tests/test_node_format_inference.py +91 -0
  1112. executorch/backends/nxp/tests/test_operator_selector.py +24 -0
  1113. executorch/backends/nxp/tests/test_per_channel_conversion.py +182 -0
  1114. executorch/backends/nxp/tests/test_qdq_clustering_conv.py +31 -0
  1115. executorch/backends/nxp/tests/test_quantizer.py +703 -0
  1116. executorch/backends/nxp/tests/test_removing_dead_code.py +65 -0
  1117. executorch/backends/nxp/tests/test_removing_nodes_with_known_outputs.py +116 -0
  1118. executorch/backends/nxp/tests/test_split_group_convolution.py +291 -0
  1119. executorch/backends/nxp/tests/use_qat.py +11 -0
  1120. executorch/backends/openvino/__init__.py +5 -0
  1121. executorch/backends/openvino/partitioner.py +238 -0
  1122. executorch/backends/openvino/preprocess.py +62 -0
  1123. executorch/backends/openvino/quantizer/__init__.py +3 -0
  1124. executorch/backends/openvino/quantizer/observers.py +186 -0
  1125. executorch/backends/openvino/quantizer/quantizer.py +595 -0
  1126. executorch/backends/openvino/tests/models/test_classification.py +38 -0
  1127. executorch/backends/openvino/tests/ops/base_openvino_op_test.py +86 -0
  1128. executorch/backends/openvino/tests/ops/test_add.py +22 -0
  1129. executorch/backends/openvino/tests/ops/test_addmm.py +28 -0
  1130. executorch/backends/openvino/tests/ops/test_arange.py +23 -0
  1131. executorch/backends/openvino/tests/ops/test_batch_norm.py +61 -0
  1132. executorch/backends/openvino/tests/ops/test_convolution.py +296 -0
  1133. executorch/backends/openvino/tests/ops/test_mean.py +90 -0
  1134. executorch/backends/openvino/tests/ops/test_permute.py +33 -0
  1135. executorch/backends/openvino/tests/ops/test_pooling.py +91 -0
  1136. executorch/backends/openvino/tests/ops/test_unary_ops.py +37 -0
  1137. executorch/backends/openvino/tests/ops/test_view.py +35 -0
  1138. executorch/backends/openvino/tests/test_runner.py +76 -0
  1139. executorch/backends/qualcomm/__init__.py +13 -0
  1140. executorch/backends/qualcomm/_passes/__init__.py +101 -0
  1141. executorch/backends/qualcomm/_passes/annotate_adaptive_avg_pool1d.py +43 -0
  1142. executorch/backends/qualcomm/_passes/annotate_quant_attrs.py +145 -0
  1143. executorch/backends/qualcomm/_passes/annotate_stack.py +44 -0
  1144. executorch/backends/qualcomm/_passes/annotate_unbind.py +42 -0
  1145. executorch/backends/qualcomm/_passes/build_quant_io.py +49 -0
  1146. executorch/backends/qualcomm/_passes/canonicalize_conv.py +204 -0
  1147. executorch/backends/qualcomm/_passes/convert_bmm_to_matmul.py +83 -0
  1148. executorch/backends/qualcomm/_passes/convert_linear_to_conv2d.py +232 -0
  1149. executorch/backends/qualcomm/_passes/convert_mha_to_sha.py +626 -0
  1150. executorch/backends/qualcomm/_passes/convert_square_to_pow.py +38 -0
  1151. executorch/backends/qualcomm/_passes/decompose_any.py +64 -0
  1152. executorch/backends/qualcomm/_passes/decompose_binary_alpha.py +61 -0
  1153. executorch/backends/qualcomm/_passes/decompose_cdist.py +69 -0
  1154. executorch/backends/qualcomm/_passes/decompose_col_im.py +121 -0
  1155. executorch/backends/qualcomm/_passes/decompose_einsum.py +51 -0
  1156. executorch/backends/qualcomm/_passes/decompose_expm1.py +46 -0
  1157. executorch/backends/qualcomm/_passes/decompose_floor_divide.py +62 -0
  1158. executorch/backends/qualcomm/_passes/decompose_glu.py +55 -0
  1159. executorch/backends/qualcomm/_passes/decompose_linalg_vector_norm.py +75 -0
  1160. executorch/backends/qualcomm/_passes/decompose_maxpool3d.py +133 -0
  1161. executorch/backends/qualcomm/_passes/decompose_minmaxdim.py +109 -0
  1162. executorch/backends/qualcomm/_passes/decompose_roll.py +78 -0
  1163. executorch/backends/qualcomm/_passes/decompose_silu.py +45 -0
  1164. executorch/backends/qualcomm/_passes/decompose_threshold.py +61 -0
  1165. executorch/backends/qualcomm/_passes/decompose_triu.py +71 -0
  1166. executorch/backends/qualcomm/_passes/decompose_wrap_with_autocast.py +79 -0
  1167. executorch/backends/qualcomm/_passes/expand_broadcast_tensor_shape.py +67 -0
  1168. executorch/backends/qualcomm/_passes/fixed_linear_keep_dim.py +95 -0
  1169. executorch/backends/qualcomm/_passes/fold_qdq.py +80 -0
  1170. executorch/backends/qualcomm/_passes/fuse_consecutive_cast.py +116 -0
  1171. executorch/backends/qualcomm/_passes/fuse_consecutive_transpose.py +109 -0
  1172. executorch/backends/qualcomm/_passes/i64_to_i32.py +184 -0
  1173. executorch/backends/qualcomm/_passes/insert_io_qdq.py +154 -0
  1174. executorch/backends/qualcomm/_passes/insert_requantize.py +111 -0
  1175. executorch/backends/qualcomm/_passes/insert_reshape_for_reduce_ops.py +59 -0
  1176. executorch/backends/qualcomm/_passes/layout_transform.py +355 -0
  1177. executorch/backends/qualcomm/_passes/lift_constant_scalar_operands.py +186 -0
  1178. executorch/backends/qualcomm/_passes/qnn_pass_manager.py +271 -0
  1179. executorch/backends/qualcomm/_passes/recompose_pad_maxpool2d.py +147 -0
  1180. executorch/backends/qualcomm/_passes/recompose_pixel_unshuffle.py +81 -0
  1181. executorch/backends/qualcomm/_passes/recompose_rms_norm.py +155 -0
  1182. executorch/backends/qualcomm/_passes/reduce_dynamic_range.py +60 -0
  1183. executorch/backends/qualcomm/_passes/remove_0d_tensor.py +36 -0
  1184. executorch/backends/qualcomm/_passes/remove_redundancy.py +73 -0
  1185. executorch/backends/qualcomm/_passes/replace_arange_args.py +48 -0
  1186. executorch/backends/qualcomm/_passes/replace_inf_values.py +60 -0
  1187. executorch/backends/qualcomm/_passes/seq_mse.py +224 -0
  1188. executorch/backends/qualcomm/_passes/tag_quant_io.py +45 -0
  1189. executorch/backends/qualcomm/_passes/utils.py +283 -0
  1190. executorch/backends/qualcomm/builders/__init__.py +221 -0
  1191. executorch/backends/qualcomm/builders/node_visitor.py +518 -0
  1192. executorch/backends/qualcomm/builders/node_visitor_manager.py +79 -0
  1193. executorch/backends/qualcomm/builders/op_abs.py +57 -0
  1194. executorch/backends/qualcomm/builders/op_adaptive_avg_pool2d.py +126 -0
  1195. executorch/backends/qualcomm/builders/op_adaptive_max_pool2d.py +151 -0
  1196. executorch/backends/qualcomm/builders/op_add.py +62 -0
  1197. executorch/backends/qualcomm/builders/op_amax.py +90 -0
  1198. executorch/backends/qualcomm/builders/op_amin.py +90 -0
  1199. executorch/backends/qualcomm/builders/op_and.py +60 -0
  1200. executorch/backends/qualcomm/builders/op_arange.py +41 -0
  1201. executorch/backends/qualcomm/builders/op_argmax.py +78 -0
  1202. executorch/backends/qualcomm/builders/op_argmin.py +78 -0
  1203. executorch/backends/qualcomm/builders/op_asin.py +56 -0
  1204. executorch/backends/qualcomm/builders/op_atan.py +55 -0
  1205. executorch/backends/qualcomm/builders/op_avg_pool2d.py +168 -0
  1206. executorch/backends/qualcomm/builders/op_avg_pool3d.py +299 -0
  1207. executorch/backends/qualcomm/builders/op_batch_norm.py +151 -0
  1208. executorch/backends/qualcomm/builders/op_binary.py +84 -0
  1209. executorch/backends/qualcomm/builders/op_bmm.py +59 -0
  1210. executorch/backends/qualcomm/builders/op_cat.py +89 -0
  1211. executorch/backends/qualcomm/builders/op_ceil.py +56 -0
  1212. executorch/backends/qualcomm/builders/op_clamp.py +80 -0
  1213. executorch/backends/qualcomm/builders/op_conv.py +263 -0
  1214. executorch/backends/qualcomm/builders/op_copy.py +112 -0
  1215. executorch/backends/qualcomm/builders/op_cos.py +56 -0
  1216. executorch/backends/qualcomm/builders/op_cum_sum.py +87 -0
  1217. executorch/backends/qualcomm/builders/op_custom_op.py +93 -0
  1218. executorch/backends/qualcomm/builders/op_depth_to_space.py +78 -0
  1219. executorch/backends/qualcomm/builders/op_dequantize.py +72 -0
  1220. executorch/backends/qualcomm/builders/op_div.py +62 -0
  1221. executorch/backends/qualcomm/builders/op_elu.py +68 -0
  1222. executorch/backends/qualcomm/builders/op_embedding.py +79 -0
  1223. executorch/backends/qualcomm/builders/op_eq.py +62 -0
  1224. executorch/backends/qualcomm/builders/op_exp.py +60 -0
  1225. executorch/backends/qualcomm/builders/op_expand.py +84 -0
  1226. executorch/backends/qualcomm/builders/op_flip.py +81 -0
  1227. executorch/backends/qualcomm/builders/op_floor.py +56 -0
  1228. executorch/backends/qualcomm/builders/op_full.py +41 -0
  1229. executorch/backends/qualcomm/builders/op_full_like.py +41 -0
  1230. executorch/backends/qualcomm/builders/op_gather.py +76 -0
  1231. executorch/backends/qualcomm/builders/op_ge.py +62 -0
  1232. executorch/backends/qualcomm/builders/op_gelu.py +57 -0
  1233. executorch/backends/qualcomm/builders/op_grid_sampler_2d.py +162 -0
  1234. executorch/backends/qualcomm/builders/op_group_norm.py +94 -0
  1235. executorch/backends/qualcomm/builders/op_gt.py +62 -0
  1236. executorch/backends/qualcomm/builders/op_hardsigmoid.py +78 -0
  1237. executorch/backends/qualcomm/builders/op_hardswish.py +57 -0
  1238. executorch/backends/qualcomm/builders/op_hardtanh.py +79 -0
  1239. executorch/backends/qualcomm/builders/op_index.py +85 -0
  1240. executorch/backends/qualcomm/builders/op_index_put.py +481 -0
  1241. executorch/backends/qualcomm/builders/op_index_select.py +81 -0
  1242. executorch/backends/qualcomm/builders/op_instance_norm.py +112 -0
  1243. executorch/backends/qualcomm/builders/op_layer_norm.py +113 -0
  1244. executorch/backends/qualcomm/builders/op_le.py +62 -0
  1245. executorch/backends/qualcomm/builders/op_linear.py +98 -0
  1246. executorch/backends/qualcomm/builders/op_log.py +58 -0
  1247. executorch/backends/qualcomm/builders/op_log_softmax.py +81 -0
  1248. executorch/backends/qualcomm/builders/op_logical_not.py +56 -0
  1249. executorch/backends/qualcomm/builders/op_lt.py +62 -0
  1250. executorch/backends/qualcomm/builders/op_matmul.py +59 -0
  1251. executorch/backends/qualcomm/builders/op_max.py +62 -0
  1252. executorch/backends/qualcomm/builders/op_max_dim.py +89 -0
  1253. executorch/backends/qualcomm/builders/op_max_pool2d.py +145 -0
  1254. executorch/backends/qualcomm/builders/op_mean_dim.py +100 -0
  1255. executorch/backends/qualcomm/builders/op_min.py +62 -0
  1256. executorch/backends/qualcomm/builders/op_min_dim.py +89 -0
  1257. executorch/backends/qualcomm/builders/op_mul.py +62 -0
  1258. executorch/backends/qualcomm/builders/op_ne.py +62 -0
  1259. executorch/backends/qualcomm/builders/op_neg.py +54 -0
  1260. executorch/backends/qualcomm/builders/op_or.py +60 -0
  1261. executorch/backends/qualcomm/builders/op_pad.py +97 -0
  1262. executorch/backends/qualcomm/builders/op_pow.py +75 -0
  1263. executorch/backends/qualcomm/builders/op_prelu.py +81 -0
  1264. executorch/backends/qualcomm/builders/op_quantize.py +73 -0
  1265. executorch/backends/qualcomm/builders/op_relu.py +58 -0
  1266. executorch/backends/qualcomm/builders/op_repeat.py +68 -0
  1267. executorch/backends/qualcomm/builders/op_reshape.py +55 -0
  1268. executorch/backends/qualcomm/builders/op_resize.py +85 -0
  1269. executorch/backends/qualcomm/builders/op_rms_norm.py +132 -0
  1270. executorch/backends/qualcomm/builders/op_round.py +58 -0
  1271. executorch/backends/qualcomm/builders/op_rsqrt.py +58 -0
  1272. executorch/backends/qualcomm/builders/op_scalar_tensor.py +51 -0
  1273. executorch/backends/qualcomm/builders/op_select_copy.py +89 -0
  1274. executorch/backends/qualcomm/builders/op_sigmoid.py +58 -0
  1275. executorch/backends/qualcomm/builders/op_sign.py +56 -0
  1276. executorch/backends/qualcomm/builders/op_sin.py +56 -0
  1277. executorch/backends/qualcomm/builders/op_skip_ops.py +55 -0
  1278. executorch/backends/qualcomm/builders/op_slice_copy.py +95 -0
  1279. executorch/backends/qualcomm/builders/op_slice_scatter.py +122 -0
  1280. executorch/backends/qualcomm/builders/op_softmax.py +79 -0
  1281. executorch/backends/qualcomm/builders/op_space_to_depth.py +78 -0
  1282. executorch/backends/qualcomm/builders/op_split_with_sizes.py +99 -0
  1283. executorch/backends/qualcomm/builders/op_sqrt.py +60 -0
  1284. executorch/backends/qualcomm/builders/op_squeeze.py +57 -0
  1285. executorch/backends/qualcomm/builders/op_stack.py +71 -0
  1286. executorch/backends/qualcomm/builders/op_sub.py +62 -0
  1287. executorch/backends/qualcomm/builders/op_sum_int_list.py +84 -0
  1288. executorch/backends/qualcomm/builders/op_tanh.py +57 -0
  1289. executorch/backends/qualcomm/builders/op_to.py +149 -0
  1290. executorch/backends/qualcomm/builders/op_topk.py +112 -0
  1291. executorch/backends/qualcomm/builders/op_transpose.py +76 -0
  1292. executorch/backends/qualcomm/builders/op_unbind.py +73 -0
  1293. executorch/backends/qualcomm/builders/op_unsqueeze.py +57 -0
  1294. executorch/backends/qualcomm/builders/op_upsample_bilinear2d.py +68 -0
  1295. executorch/backends/qualcomm/builders/op_upsample_nearest2d.py +68 -0
  1296. executorch/backends/qualcomm/builders/op_where.py +82 -0
  1297. executorch/backends/qualcomm/builders/op_xor.py +60 -0
  1298. executorch/backends/qualcomm/builders/qnn_constants.py +630 -0
  1299. executorch/backends/qualcomm/builders/utils.py +160 -0
  1300. executorch/backends/qualcomm/debugger/format_outputs.py +221 -0
  1301. executorch/backends/qualcomm/debugger/metrics_evaluator.py +90 -0
  1302. executorch/backends/qualcomm/debugger/qnn_intermediate_debugger.py +328 -0
  1303. executorch/backends/qualcomm/debugger/utils.py +435 -0
  1304. executorch/backends/qualcomm/partition/common_defs.py +44 -0
  1305. executorch/backends/qualcomm/partition/qnn_partitioner.py +225 -0
  1306. executorch/backends/qualcomm/partition/utils.py +75 -0
  1307. executorch/backends/qualcomm/qnn_preprocess.py +218 -0
  1308. executorch/backends/qualcomm/quantizer/annotators.py +1501 -0
  1309. executorch/backends/qualcomm/quantizer/custom_annotation.py +499 -0
  1310. executorch/backends/qualcomm/quantizer/observers/concat_observer.py +74 -0
  1311. executorch/backends/qualcomm/quantizer/observers/per_block_param_observer.py +145 -0
  1312. executorch/backends/qualcomm/quantizer/observers/per_channel_param_observer.py +113 -0
  1313. executorch/backends/qualcomm/quantizer/qconfig.py +802 -0
  1314. executorch/backends/qualcomm/quantizer/quant_recipe.py +402 -0
  1315. executorch/backends/qualcomm/quantizer/quantizer.py +507 -0
  1316. executorch/backends/qualcomm/recipes/__init__.py +16 -0
  1317. executorch/backends/qualcomm/recipes/qnn_recipe_provider.py +179 -0
  1318. executorch/backends/qualcomm/recipes/qnn_recipe_types.py +28 -0
  1319. executorch/backends/qualcomm/scripts/download_qnn_sdk.py +642 -0
  1320. executorch/backends/qualcomm/serialization/qc_compiler_spec.fbs +299 -0
  1321. executorch/backends/qualcomm/serialization/qc_schema.py +240 -0
  1322. executorch/backends/qualcomm/serialization/qc_schema_serialize.py +59 -0
  1323. executorch/backends/qualcomm/tests/models.py +2393 -0
  1324. executorch/backends/qualcomm/tests/test_passes.py +154 -0
  1325. executorch/backends/qualcomm/tests/test_qnn_delegate.py +8454 -0
  1326. executorch/backends/qualcomm/tests/tester.py +113 -0
  1327. executorch/backends/qualcomm/tests/utils.py +770 -0
  1328. executorch/backends/qualcomm/utils/constants.py +49 -0
  1329. executorch/backends/qualcomm/utils/qnn_manager_lifecycle.py +88 -0
  1330. executorch/backends/qualcomm/utils/utils.py +1302 -0
  1331. executorch/backends/samsung/_passes/annotate_qparams.py +201 -0
  1332. executorch/backends/samsung/_passes/annotate_scalar_parameters.py +65 -0
  1333. executorch/backends/samsung/_passes/conv1d_to_conv2d.py +97 -0
  1334. executorch/backends/samsung/_passes/customized_constant_prop.py +40 -0
  1335. executorch/backends/samsung/_passes/fold_qdq.py +36 -0
  1336. executorch/backends/samsung/_passes/fuse_conv_act.py +77 -0
  1337. executorch/backends/samsung/_passes/insert_qdq.py +164 -0
  1338. executorch/backends/samsung/_passes/remove_useless_ops.py +87 -0
  1339. executorch/backends/samsung/_passes/replace_scalar_ops.py +46 -0
  1340. executorch/backends/samsung/_passes/utils.py +15 -0
  1341. executorch/backends/samsung/builders/__init__.py +97 -0
  1342. executorch/backends/samsung/builders/node_visitor.py +116 -0
  1343. executorch/backends/samsung/builders/op_add.py +41 -0
  1344. executorch/backends/samsung/builders/op_avg_pool2d.py +70 -0
  1345. executorch/backends/samsung/builders/op_batch_norm.py +56 -0
  1346. executorch/backends/samsung/builders/op_bmm.py +43 -0
  1347. executorch/backends/samsung/builders/op_cat.py +50 -0
  1348. executorch/backends/samsung/builders/op_clamp.py +47 -0
  1349. executorch/backends/samsung/builders/op_constant_pad_nd.py +56 -0
  1350. executorch/backends/samsung/builders/op_conv2d.py +89 -0
  1351. executorch/backends/samsung/builders/op_dequantize.py +19 -0
  1352. executorch/backends/samsung/builders/op_div.py +42 -0
  1353. executorch/backends/samsung/builders/op_embedding.py +41 -0
  1354. executorch/backends/samsung/builders/op_expand_copy.py +77 -0
  1355. executorch/backends/samsung/builders/op_gelu.py +40 -0
  1356. executorch/backends/samsung/builders/op_getitem.py +32 -0
  1357. executorch/backends/samsung/builders/op_hardsigmoid.py +35 -0
  1358. executorch/backends/samsung/builders/op_hardswish.py +35 -0
  1359. executorch/backends/samsung/builders/op_hardtanh.py +42 -0
  1360. executorch/backends/samsung/builders/op_layer_norm.py +53 -0
  1361. executorch/backends/samsung/builders/op_leaky_relu.py +58 -0
  1362. executorch/backends/samsung/builders/op_linear.py +51 -0
  1363. executorch/backends/samsung/builders/op_log_softmax.py +39 -0
  1364. executorch/backends/samsung/builders/op_max_pool2d.py +96 -0
  1365. executorch/backends/samsung/builders/op_maximum.py +37 -0
  1366. executorch/backends/samsung/builders/op_mean_dim.py +49 -0
  1367. executorch/backends/samsung/builders/op_minimum.py +39 -0
  1368. executorch/backends/samsung/builders/op_mul.py +43 -0
  1369. executorch/backends/samsung/builders/op_permute.py +37 -0
  1370. executorch/backends/samsung/builders/op_pixel_shuffle.py +38 -0
  1371. executorch/backends/samsung/builders/op_quantize.py +60 -0
  1372. executorch/backends/samsung/builders/op_relu.py +37 -0
  1373. executorch/backends/samsung/builders/op_reshape.py +34 -0
  1374. executorch/backends/samsung/builders/op_rsqrt.py +35 -0
  1375. executorch/backends/samsung/builders/op_select.py +52 -0
  1376. executorch/backends/samsung/builders/op_slice_copy.py +59 -0
  1377. executorch/backends/samsung/builders/op_softmax.py +39 -0
  1378. executorch/backends/samsung/builders/op_sqrt.py +35 -0
  1379. executorch/backends/samsung/builders/op_squeeze.py +37 -0
  1380. executorch/backends/samsung/builders/op_sub.py +39 -0
  1381. executorch/backends/samsung/builders/op_to_copy.py +44 -0
  1382. executorch/backends/samsung/builders/op_unsqueeze.py +35 -0
  1383. executorch/backends/samsung/builders/op_upsample_bilinear2d.py +53 -0
  1384. executorch/backends/samsung/builders/op_upsample_nearest2d.py +52 -0
  1385. executorch/backends/samsung/builders/utils.py +64 -0
  1386. executorch/backends/samsung/enn_preprocess.py +110 -0
  1387. executorch/backends/samsung/partition/enn_partitioner.py +134 -0
  1388. executorch/backends/samsung/quantizer/__init__.py +10 -0
  1389. executorch/backends/samsung/quantizer/annotator.py +871 -0
  1390. executorch/backends/samsung/quantizer/qconfig.py +174 -0
  1391. executorch/backends/samsung/quantizer/quantizer.py +65 -0
  1392. executorch/backends/samsung/serialization/compile_options.py +77 -0
  1393. executorch/backends/samsung/serialization/compile_options_def.fbs +23 -0
  1394. executorch/backends/samsung/serialization/enn_graph_schema.py +141 -0
  1395. executorch/backends/samsung/test/models/test_deeplab_v3.py +28 -0
  1396. executorch/backends/samsung/test/models/test_edsr.py +30 -0
  1397. executorch/backends/samsung/test/models/test_inception_v3.py +30 -0
  1398. executorch/backends/samsung/test/models/test_inception_v4.py +30 -0
  1399. executorch/backends/samsung/test/models/test_mobilenet_v2.py +28 -0
  1400. executorch/backends/samsung/test/models/test_mobilenet_v3.py +33 -0
  1401. executorch/backends/samsung/test/models/test_resnet18.py +30 -0
  1402. executorch/backends/samsung/test/models/test_resnet50.py +30 -0
  1403. executorch/backends/samsung/test/models/test_torchvision_vit.py +32 -0
  1404. executorch/backends/samsung/test/models/test_wav2letter.py +28 -0
  1405. executorch/backends/samsung/test/ops/test_add.py +63 -0
  1406. executorch/backends/samsung/test/ops/test_avg_pool2d.py +66 -0
  1407. executorch/backends/samsung/test/ops/test_batch_norm.py +52 -0
  1408. executorch/backends/samsung/test/ops/test_bmm.py +50 -0
  1409. executorch/backends/samsung/test/ops/test_cat.py +49 -0
  1410. executorch/backends/samsung/test/ops/test_clamp.py +50 -0
  1411. executorch/backends/samsung/test/ops/test_constant_pad_nd.py +46 -0
  1412. executorch/backends/samsung/test/ops/test_conv2d.py +95 -0
  1413. executorch/backends/samsung/test/ops/test_div.py +50 -0
  1414. executorch/backends/samsung/test/ops/test_embedding.py +46 -0
  1415. executorch/backends/samsung/test/ops/test_expand_copy.py +46 -0
  1416. executorch/backends/samsung/test/ops/test_gelu.py +65 -0
  1417. executorch/backends/samsung/test/ops/test_leaky_relu.py +47 -0
  1418. executorch/backends/samsung/test/ops/test_linear.py +48 -0
  1419. executorch/backends/samsung/test/ops/test_log_softmax.py +46 -0
  1420. executorch/backends/samsung/test/ops/test_max_pool2d.py +72 -0
  1421. executorch/backends/samsung/test/ops/test_mean_dim.py +51 -0
  1422. executorch/backends/samsung/test/ops/test_minimum.py +45 -0
  1423. executorch/backends/samsung/test/ops/test_mul.py +63 -0
  1424. executorch/backends/samsung/test/ops/test_permute.py +55 -0
  1425. executorch/backends/samsung/test/ops/test_pixel_shuffle.py +49 -0
  1426. executorch/backends/samsung/test/ops/test_relu.py +65 -0
  1427. executorch/backends/samsung/test/ops/test_reshape.py +46 -0
  1428. executorch/backends/samsung/test/ops/test_rsqrt.py +46 -0
  1429. executorch/backends/samsung/test/ops/test_select.py +52 -0
  1430. executorch/backends/samsung/test/ops/test_slice_copy.py +46 -0
  1431. executorch/backends/samsung/test/ops/test_softmax.py +46 -0
  1432. executorch/backends/samsung/test/ops/test_sqrt.py +46 -0
  1433. executorch/backends/samsung/test/ops/test_squeeze.py +47 -0
  1434. executorch/backends/samsung/test/ops/test_sub.py +63 -0
  1435. executorch/backends/samsung/test/ops/test_to_copy.py +49 -0
  1436. executorch/backends/samsung/test/ops/test_unsqueeze.py +47 -0
  1437. executorch/backends/samsung/test/ops/test_upsample_bilinear2d.py +54 -0
  1438. executorch/backends/samsung/test/ops/test_upsample_nearest2d.py +53 -0
  1439. executorch/backends/samsung/test/tester/__init__.py +11 -0
  1440. executorch/backends/samsung/test/tester/samsung_tester.py +125 -0
  1441. executorch/backends/samsung/test/utils/__init__.py +11 -0
  1442. executorch/backends/samsung/test/utils/runtime_executor.py +186 -0
  1443. executorch/backends/samsung/utils/constants.py +45 -0
  1444. executorch/backends/samsung/utils/export_utils.py +93 -0
  1445. executorch/backends/samsung/utils/utils.py +125 -0
  1446. executorch/backends/test/facto/__init__.py +0 -0
  1447. executorch/backends/test/facto/facto_specs.py +59 -0
  1448. executorch/backends/test/facto/test_facto.py +301 -0
  1449. executorch/backends/test/harness/__init__.py +3 -0
  1450. executorch/backends/test/harness/error_statistics.py +99 -0
  1451. executorch/backends/test/harness/stages/__init__.py +23 -0
  1452. executorch/backends/test/harness/stages/export.py +32 -0
  1453. executorch/backends/test/harness/stages/partition.py +26 -0
  1454. executorch/backends/test/harness/stages/quantize.py +128 -0
  1455. executorch/backends/test/harness/stages/run_passes.py +66 -0
  1456. executorch/backends/test/harness/stages/serialize.py +59 -0
  1457. executorch/backends/test/harness/stages/stage.py +91 -0
  1458. executorch/backends/test/harness/stages/to_edge.py +27 -0
  1459. executorch/backends/test/harness/stages/to_edge_transform_and_lower.py +50 -0
  1460. executorch/backends/test/harness/stages/to_executorch.py +54 -0
  1461. executorch/backends/test/harness/tester.py +478 -0
  1462. executorch/backends/test/harness/tests/test_error_statistics.py +67 -0
  1463. executorch/backends/test/suite/__init__.py +74 -0
  1464. executorch/backends/test/suite/conftest.py +182 -0
  1465. executorch/backends/test/suite/context.py +34 -0
  1466. executorch/backends/test/suite/discovery.py +100 -0
  1467. executorch/backends/test/suite/flow.py +150 -0
  1468. executorch/backends/test/suite/flows/__init__.py +7 -0
  1469. executorch/backends/test/suite/flows/arm.py +92 -0
  1470. executorch/backends/test/suite/flows/coreml.py +31 -0
  1471. executorch/backends/test/suite/flows/portable.py +19 -0
  1472. executorch/backends/test/suite/flows/qualcomm.py +61 -0
  1473. executorch/backends/test/suite/flows/vulkan.py +44 -0
  1474. executorch/backends/test/suite/flows/xnnpack.py +79 -0
  1475. executorch/backends/test/suite/generate_markdown_summary.py +249 -0
  1476. executorch/backends/test/suite/generate_markdown_summary_json.py +229 -0
  1477. executorch/backends/test/suite/models/__init__.py +7 -0
  1478. executorch/backends/test/suite/models/test_torchaudio.py +110 -0
  1479. executorch/backends/test/suite/models/test_torchvision.py +194 -0
  1480. executorch/backends/test/suite/operators/__init__.py +108 -0
  1481. executorch/backends/test/suite/operators/test_abs.py +60 -0
  1482. executorch/backends/test/suite/operators/test_adaptive_avgpool1d.py +102 -0
  1483. executorch/backends/test/suite/operators/test_adaptive_avgpool2d.py +112 -0
  1484. executorch/backends/test/suite/operators/test_adaptive_avgpool3d.py +112 -0
  1485. executorch/backends/test/suite/operators/test_adaptive_maxpool1d.py +125 -0
  1486. executorch/backends/test/suite/operators/test_adaptive_maxpool2d.py +135 -0
  1487. executorch/backends/test/suite/operators/test_adaptive_maxpool3d.py +135 -0
  1488. executorch/backends/test/suite/operators/test_add.py +76 -0
  1489. executorch/backends/test/suite/operators/test_amax.py +257 -0
  1490. executorch/backends/test/suite/operators/test_amin.py +259 -0
  1491. executorch/backends/test/suite/operators/test_argmax.py +201 -0
  1492. executorch/backends/test/suite/operators/test_argmin.py +201 -0
  1493. executorch/backends/test/suite/operators/test_avgpool1d.py +155 -0
  1494. executorch/backends/test/suite/operators/test_avgpool2d.py +168 -0
  1495. executorch/backends/test/suite/operators/test_avgpool3d.py +163 -0
  1496. executorch/backends/test/suite/operators/test_cat.py +176 -0
  1497. executorch/backends/test/suite/operators/test_ceil.py +60 -0
  1498. executorch/backends/test/suite/operators/test_clamp.py +86 -0
  1499. executorch/backends/test/suite/operators/test_conv1d.py +144 -0
  1500. executorch/backends/test/suite/operators/test_conv2d.py +167 -0
  1501. executorch/backends/test/suite/operators/test_conv3d.py +162 -0
  1502. executorch/backends/test/suite/operators/test_convtranspose1d.py +144 -0
  1503. executorch/backends/test/suite/operators/test_convtranspose2d.py +171 -0
  1504. executorch/backends/test/suite/operators/test_convtranspose3d.py +166 -0
  1505. executorch/backends/test/suite/operators/test_div.py +110 -0
  1506. executorch/backends/test/suite/operators/test_elu.py +49 -0
  1507. executorch/backends/test/suite/operators/test_embedding.py +89 -0
  1508. executorch/backends/test/suite/operators/test_embedding_bag.py +118 -0
  1509. executorch/backends/test/suite/operators/test_exp.py +65 -0
  1510. executorch/backends/test/suite/operators/test_expand.py +122 -0
  1511. executorch/backends/test/suite/operators/test_floor.py +57 -0
  1512. executorch/backends/test/suite/operators/test_floor_divide.py +216 -0
  1513. executorch/backends/test/suite/operators/test_gelu.py +52 -0
  1514. executorch/backends/test/suite/operators/test_glu.py +52 -0
  1515. executorch/backends/test/suite/operators/test_hardsigmoid.py +50 -0
  1516. executorch/backends/test/suite/operators/test_hardswish.py +50 -0
  1517. executorch/backends/test/suite/operators/test_hardtanh.py +57 -0
  1518. executorch/backends/test/suite/operators/test_index_put.py +455 -0
  1519. executorch/backends/test/suite/operators/test_index_select.py +128 -0
  1520. executorch/backends/test/suite/operators/test_leaky_relu.py +56 -0
  1521. executorch/backends/test/suite/operators/test_linear.py +124 -0
  1522. executorch/backends/test/suite/operators/test_log.py +80 -0
  1523. executorch/backends/test/suite/operators/test_log10.py +60 -0
  1524. executorch/backends/test/suite/operators/test_log1p.py +60 -0
  1525. executorch/backends/test/suite/operators/test_log2.py +60 -0
  1526. executorch/backends/test/suite/operators/test_logsigmoid.py +40 -0
  1527. executorch/backends/test/suite/operators/test_lstm.py +247 -0
  1528. executorch/backends/test/suite/operators/test_masked_fill.py +101 -0
  1529. executorch/backends/test/suite/operators/test_maxpool1d.py +185 -0
  1530. executorch/backends/test/suite/operators/test_maxpool2d.py +191 -0
  1531. executorch/backends/test/suite/operators/test_maxpool3d.py +189 -0
  1532. executorch/backends/test/suite/operators/test_mean.py +305 -0
  1533. executorch/backends/test/suite/operators/test_median.py +188 -0
  1534. executorch/backends/test/suite/operators/test_mul.py +66 -0
  1535. executorch/backends/test/suite/operators/test_neg.py +70 -0
  1536. executorch/backends/test/suite/operators/test_permute.py +109 -0
  1537. executorch/backends/test/suite/operators/test_pow.py +149 -0
  1538. executorch/backends/test/suite/operators/test_prelu.py +55 -0
  1539. executorch/backends/test/suite/operators/test_relu.py +45 -0
  1540. executorch/backends/test/suite/operators/test_reshape.py +82 -0
  1541. executorch/backends/test/suite/operators/test_round.py +129 -0
  1542. executorch/backends/test/suite/operators/test_rsqrt.py +76 -0
  1543. executorch/backends/test/suite/operators/test_select.py +89 -0
  1544. executorch/backends/test/suite/operators/test_sigmoid.py +40 -0
  1545. executorch/backends/test/suite/operators/test_silu.py +50 -0
  1546. executorch/backends/test/suite/operators/test_slice.py +146 -0
  1547. executorch/backends/test/suite/operators/test_split.py +119 -0
  1548. executorch/backends/test/suite/operators/test_sqrt.py +71 -0
  1549. executorch/backends/test/suite/operators/test_square.py +67 -0
  1550. executorch/backends/test/suite/operators/test_squeeze.py +86 -0
  1551. executorch/backends/test/suite/operators/test_stack.py +114 -0
  1552. executorch/backends/test/suite/operators/test_sub.py +84 -0
  1553. executorch/backends/test/suite/operators/test_tanh.py +40 -0
  1554. executorch/backends/test/suite/operators/test_threshold.py +71 -0
  1555. executorch/backends/test/suite/operators/test_transpose.py +143 -0
  1556. executorch/backends/test/suite/operators/test_trunc.py +73 -0
  1557. executorch/backends/test/suite/operators/test_unsqueeze.py +110 -0
  1558. executorch/backends/test/suite/operators/test_upsample_bilinear2d.py +273 -0
  1559. executorch/backends/test/suite/operators/test_upsample_nearest2d.py +137 -0
  1560. executorch/backends/test/suite/operators/test_view.py +82 -0
  1561. executorch/backends/test/suite/reporting.py +421 -0
  1562. executorch/backends/test/suite/runner.py +328 -0
  1563. executorch/backends/test/suite/tests/__init__.py +0 -0
  1564. executorch/backends/test/suite/tests/test_reporting.py +144 -0
  1565. executorch/backends/transforms/__init__.py +12 -0
  1566. executorch/backends/transforms/addmm_mm_to_linear.py +170 -0
  1567. executorch/backends/transforms/convert_dtype_pass.py +103 -0
  1568. executorch/backends/transforms/decompose_sdpa.py +120 -0
  1569. executorch/backends/transforms/duplicate_dynamic_quant_chain.py +195 -0
  1570. executorch/backends/transforms/fuse_batch_norm_with_conv.py +161 -0
  1571. executorch/backends/transforms/fuse_conv_with_clamp.py +77 -0
  1572. executorch/backends/transforms/fuse_view_copy.py +116 -0
  1573. executorch/backends/transforms/mean_to_sum_div.py +41 -0
  1574. executorch/backends/transforms/rank_0_to_rank_1.py +18 -0
  1575. executorch/backends/transforms/remove_clone_ops.py +109 -0
  1576. executorch/backends/transforms/remove_getitem_op.py +109 -0
  1577. executorch/backends/transforms/replace_scalar_with_tensor.py +93 -0
  1578. executorch/backends/transforms/test/test_create_delete_constant_placeholder.py +122 -0
  1579. executorch/backends/transforms/test/test_create_mutable_buffer.py +503 -0
  1580. executorch/backends/transforms/test/test_duplicate_dynamic_quant_chain.py +184 -0
  1581. executorch/backends/transforms/test/test_rank_0_to_rank_1.py +32 -0
  1582. executorch/backends/transforms/test/test_remove_clone_ops.py +197 -0
  1583. executorch/backends/transforms/utils.py +377 -0
  1584. executorch/backends/transforms/view_copy_to_squeeze_unsqueeze.py +135 -0
  1585. executorch/backends/vulkan/__init__.py +14 -0
  1586. executorch/backends/vulkan/_passes/__init__.py +37 -0
  1587. executorch/backends/vulkan/_passes/fold_qdq.py +40 -0
  1588. executorch/backends/vulkan/_passes/fuse_patterns.py +34 -0
  1589. executorch/backends/vulkan/_passes/fuse_quantized_ops.py +236 -0
  1590. executorch/backends/vulkan/_passes/insert_prepack_nodes.py +83 -0
  1591. executorch/backends/vulkan/_passes/remove_asserts.py +52 -0
  1592. executorch/backends/vulkan/_passes/remove_redundant_ops.py +72 -0
  1593. executorch/backends/vulkan/_passes/squeeze_unsqueeze_inputs.py +93 -0
  1594. executorch/backends/vulkan/_passes/tag_memory_meta_pass.py +485 -0
  1595. executorch/backends/vulkan/custom_ops_lib.py +597 -0
  1596. executorch/backends/vulkan/op_registry.py +885 -0
  1597. executorch/backends/vulkan/partitioner/vulkan_partitioner.py +401 -0
  1598. executorch/backends/vulkan/patterns/__init__.py +139 -0
  1599. executorch/backends/vulkan/patterns/pattern_registry.py +116 -0
  1600. executorch/backends/vulkan/patterns/quantized_binary.py +161 -0
  1601. executorch/backends/vulkan/patterns/quantized_convolution.py +260 -0
  1602. executorch/backends/vulkan/patterns/quantized_linear.py +500 -0
  1603. executorch/backends/vulkan/patterns/rope.py +173 -0
  1604. executorch/backends/vulkan/patterns/sdpa.py +155 -0
  1605. executorch/backends/vulkan/patterns/select_as_symint.py +104 -0
  1606. executorch/backends/vulkan/quantizer/vulkan_quantizer.py +183 -0
  1607. executorch/backends/vulkan/quantizer/vulkan_quantizer_utils.py +223 -0
  1608. executorch/backends/vulkan/runtime/gen_vulkan_spv.py +1517 -0
  1609. executorch/backends/vulkan/runtime/graph/ops/glsl/addmm_naive_buffer.yaml +19 -0
  1610. executorch/backends/vulkan/runtime/graph/ops/glsl/addmm_naive_texture3d.yaml +24 -0
  1611. executorch/backends/vulkan/runtime/graph/ops/glsl/addmm_optimized.yaml +43 -0
  1612. executorch/backends/vulkan/runtime/graph/ops/glsl/arange.yaml +19 -0
  1613. executorch/backends/vulkan/runtime/graph/ops/glsl/avg_pool2d.yaml +18 -0
  1614. executorch/backends/vulkan/runtime/graph/ops/glsl/batchnorm.yaml +11 -0
  1615. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_op.yaml +118 -0
  1616. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_q8ta_q8ta_q8to.yaml +19 -0
  1617. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_buffer.yaml +20 -0
  1618. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_texture.yaml +20 -0
  1619. executorch/backends/vulkan/runtime/graph/ops/glsl/bitw8_image_to_nchw_nobitw8buffer.yaml +22 -0
  1620. executorch/backends/vulkan/runtime/graph/ops/glsl/buffer_to_buffer.yaml +20 -0
  1621. executorch/backends/vulkan/runtime/graph/ops/glsl/buffer_to_nchw.yaml +25 -0
  1622. executorch/backends/vulkan/runtime/graph/ops/glsl/choose_qparams_per_row.yaml +19 -0
  1623. executorch/backends/vulkan/runtime/graph/ops/glsl/clone.yaml +13 -0
  1624. executorch/backends/vulkan/runtime/graph/ops/glsl/col2im.yaml +19 -0
  1625. executorch/backends/vulkan/runtime/graph/ops/glsl/concat_buffer.yaml +15 -0
  1626. executorch/backends/vulkan/runtime/graph/ops/glsl/concat_texture.yaml +15 -0
  1627. executorch/backends/vulkan/runtime/graph/ops/glsl/conv1d.yaml +19 -0
  1628. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d.yaml +19 -0
  1629. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.yaml +19 -0
  1630. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.yaml +27 -0
  1631. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_prepack_weights.yaml +15 -0
  1632. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.yaml +20 -0
  1633. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_sned_output_tile.yaml +25 -0
  1634. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_prepack_weights.yaml +15 -0
  1635. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.yaml +21 -0
  1636. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_q8ta_q8csw_q8to_tiled.yaml +20 -0
  1637. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_s1p0.yaml +19 -0
  1638. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8csw_linear_tiled.yaml +22 -0
  1639. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_linear_tiled.yaml +22 -0
  1640. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to.yaml +21 -0
  1641. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to_linear_tiled.yaml +21 -0
  1642. executorch/backends/vulkan/runtime/graph/ops/glsl/conv_transpose2d.yaml +19 -0
  1643. executorch/backends/vulkan/runtime/graph/ops/glsl/conv_transpose2d_prepack_weights.yaml +16 -0
  1644. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_buffer.yaml +16 -0
  1645. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_legacy.yaml +12 -0
  1646. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_texture.yaml +15 -0
  1647. executorch/backends/vulkan/runtime/graph/ops/glsl/expand_buffer.yaml +11 -0
  1648. executorch/backends/vulkan/runtime/graph/ops/glsl/flip.yaml +14 -0
  1649. executorch/backends/vulkan/runtime/graph/ops/glsl/full.yaml +20 -0
  1650. executorch/backends/vulkan/runtime/graph/ops/glsl/gather_buffer.yaml +18 -0
  1651. executorch/backends/vulkan/runtime/graph/ops/glsl/gather_texture.yaml +17 -0
  1652. executorch/backends/vulkan/runtime/graph/ops/glsl/grid_priors.yaml +12 -0
  1653. executorch/backends/vulkan/runtime/graph/ops/glsl/group_norm_reduce_texture.yaml +15 -0
  1654. executorch/backends/vulkan/runtime/graph/ops/glsl/group_norm_texture.yaml +15 -0
  1655. executorch/backends/vulkan/runtime/graph/ops/glsl/im2col.yaml +18 -0
  1656. executorch/backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.yaml +15 -0
  1657. executorch/backends/vulkan/runtime/graph/ops/glsl/image_to_nchw.yaml +38 -0
  1658. executorch/backends/vulkan/runtime/graph/ops/glsl/index_select.yaml +13 -0
  1659. executorch/backends/vulkan/runtime/graph/ops/glsl/index_select_channel.yaml +13 -0
  1660. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_dq8ca_q4gsw_tiled.yaml +29 -0
  1661. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_coop.yaml +43 -0
  1662. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_tiled.yaml +28 -0
  1663. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q8csw_tiled.yaml +28 -0
  1664. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q8ta_q8csw_tiled.yaml +24 -0
  1665. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw.yaml +23 -0
  1666. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_coop.yaml +38 -0
  1667. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_tiled.yaml +54 -0
  1668. executorch/backends/vulkan/runtime/graph/ops/glsl/max_pool2d.yaml +17 -0
  1669. executorch/backends/vulkan/runtime/graph/ops/glsl/native_layer_norm.yaml +16 -0
  1670. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_bitw8_image_nobitw8buffer.yaml +22 -0
  1671. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_buffer.yaml +25 -0
  1672. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_image.yaml +37 -0
  1673. executorch/backends/vulkan/runtime/graph/ops/glsl/no_op.yaml +24 -0
  1674. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_int4_linear_weight_transposed_interleaved.yaml +19 -0
  1675. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q4_linear_weight.yaml +14 -0
  1676. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_conv2d_dw_weights.yaml +15 -0
  1677. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_conv2d_weights.yaml +15 -0
  1678. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_linear_weight.yaml +14 -0
  1679. executorch/backends/vulkan/runtime/graph/ops/glsl/pad_channel.yaml +14 -0
  1680. executorch/backends/vulkan/runtime/graph/ops/glsl/pad_height_width.yaml +14 -0
  1681. executorch/backends/vulkan/runtime/graph/ops/glsl/permute_buffer.yaml +11 -0
  1682. executorch/backends/vulkan/runtime/graph/ops/glsl/permute_texture.yaml +11 -0
  1683. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4h4w.yaml +25 -0
  1684. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4h4w_with_group_sums.yaml +30 -0
  1685. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4w4c.yaml +23 -0
  1686. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_im2col.yaml +18 -0
  1687. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce.yaml +29 -0
  1688. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce2d.yaml +29 -0
  1689. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce_per_row_buffer.yaml +42 -0
  1690. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat.yaml +14 -0
  1691. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat_channel.yaml +12 -0
  1692. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat_interleave.yaml +12 -0
  1693. executorch/backends/vulkan/runtime/graph/ops/glsl/rotary_embedding.yaml +13 -0
  1694. executorch/backends/vulkan/runtime/graph/ops/glsl/scalar_tensor.yaml +27 -0
  1695. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_attn_weights_softmax.yaml +19 -0
  1696. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_attn_weights_coop.yaml +25 -0
  1697. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_attn_weights_tiled.yaml +26 -0
  1698. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_out_coop.yaml +25 -0
  1699. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_out_tiled.yaml +26 -0
  1700. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_kv_cache_update.yaml +23 -0
  1701. executorch/backends/vulkan/runtime/graph/ops/glsl/set_zero.yaml +8 -0
  1702. executorch/backends/vulkan/runtime/graph/ops/glsl/softmax.yaml +21 -0
  1703. executorch/backends/vulkan/runtime/graph/ops/glsl/split_buffer.yaml +18 -0
  1704. executorch/backends/vulkan/runtime/graph/ops/glsl/split_texture.yaml +17 -0
  1705. executorch/backends/vulkan/runtime/graph/ops/glsl/tan.yaml +13 -0
  1706. executorch/backends/vulkan/runtime/graph/ops/glsl/transfer_buffer.yaml +22 -0
  1707. executorch/backends/vulkan/runtime/graph/ops/glsl/transfer_texture.yaml +22 -0
  1708. executorch/backends/vulkan/runtime/graph/ops/glsl/unary_op.yaml +48 -0
  1709. executorch/backends/vulkan/runtime/graph/ops/glsl/unpack_4w4c_and_dequantize.yaml +23 -0
  1710. executorch/backends/vulkan/runtime/graph/ops/glsl/update_concat_offset.yaml +13 -0
  1711. executorch/backends/vulkan/runtime/graph/ops/glsl/upsample_2d.yaml +19 -0
  1712. executorch/backends/vulkan/runtime/graph/ops/glsl/var_buffer.yaml +16 -0
  1713. executorch/backends/vulkan/runtime/graph/ops/glsl/var_texture3d.yaml +16 -0
  1714. executorch/backends/vulkan/runtime/graph/ops/glsl/view.yaml +13 -0
  1715. executorch/backends/vulkan/runtime/graph/ops/glsl/view_buffer.yaml +20 -0
  1716. executorch/backends/vulkan/runtime/graph/ops/glsl/view_convert_buffer.yaml +22 -0
  1717. executorch/backends/vulkan/runtime/graph/ops/glsl/where.yaml +12 -0
  1718. executorch/backends/vulkan/serialization/schema.fbs +158 -0
  1719. executorch/backends/vulkan/serialization/vulkan_graph_builder.py +505 -0
  1720. executorch/backends/vulkan/serialization/vulkan_graph_schema.py +161 -0
  1721. executorch/backends/vulkan/serialization/vulkan_graph_serialize.py +274 -0
  1722. executorch/backends/vulkan/test/custom_ops/glsl/add.yaml +29 -0
  1723. executorch/backends/vulkan/test/glsl/all_shaders.yaml +68 -0
  1724. executorch/backends/vulkan/test/glsl/dynamic_dispatch_test.yaml +7 -0
  1725. executorch/backends/vulkan/test/glsl/reference_matmul_common.yaml +9 -0
  1726. executorch/backends/vulkan/test/op_tests/cases.py +2026 -0
  1727. executorch/backends/vulkan/test/op_tests/generate_op_benchmarks.py +88 -0
  1728. executorch/backends/vulkan/test/op_tests/generate_op_correctness_tests.py +90 -0
  1729. executorch/backends/vulkan/test/op_tests/utils/aten_types.py +31 -0
  1730. executorch/backends/vulkan/test/op_tests/utils/gen_benchmark_vk.py +350 -0
  1731. executorch/backends/vulkan/test/op_tests/utils/gen_computegraph.py +788 -0
  1732. executorch/backends/vulkan/test/op_tests/utils/gen_correctness_base.py +421 -0
  1733. executorch/backends/vulkan/test/op_tests/utils/gen_correctness_vk.py +164 -0
  1734. executorch/backends/vulkan/test/op_tests/utils/test_suite.py +53 -0
  1735. executorch/backends/vulkan/test/test_serialization.py +135 -0
  1736. executorch/backends/vulkan/test/test_vulkan_delegate.py +2696 -0
  1737. executorch/backends/vulkan/test/test_vulkan_delegate_header.py +104 -0
  1738. executorch/backends/vulkan/test/test_vulkan_passes.py +193 -0
  1739. executorch/backends/vulkan/test/tester.py +91 -0
  1740. executorch/backends/vulkan/test/utils.py +1111 -0
  1741. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/apiconventions.py +21 -0
  1742. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/base_generator.py +937 -0
  1743. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/cgenerator.py +549 -0
  1744. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/generator.py +1424 -0
  1745. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/parse_dependency.py +404 -0
  1746. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/reg.py +1885 -0
  1747. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/spec_tools/conventions.py +562 -0
  1748. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/spec_tools/util.py +58 -0
  1749. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/stripAPI.py +42 -0
  1750. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/vkconventions.py +314 -0
  1751. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/vulkan_object.py +484 -0
  1752. executorch/backends/vulkan/third-party/VulkanMemoryAllocator/tools/GpuMemDumpVis/GpuMemDumpVis.py +333 -0
  1753. executorch/backends/vulkan/third-party/volk/generate.py +212 -0
  1754. executorch/backends/vulkan/tools/gpuinfo/glsl/buf_bandwidth.yaml +18 -0
  1755. executorch/backends/vulkan/tools/gpuinfo/glsl/buf_cacheline_size.yaml +12 -0
  1756. executorch/backends/vulkan/tools/gpuinfo/glsl/reg_count.yaml +16 -0
  1757. executorch/backends/vulkan/tools/gpuinfo/glsl/tex_bandwidth.yaml +15 -0
  1758. executorch/backends/vulkan/tools/gpuinfo/glsl/tex_cacheline_concurr.yaml +14 -0
  1759. executorch/backends/vulkan/tools/gpuinfo/glsl/warp_size.yaml +16 -0
  1760. executorch/backends/vulkan/utils.py +1474 -0
  1761. executorch/backends/vulkan/vulkan_preprocess.py +249 -0
  1762. executorch/backends/xnnpack/__init__.py +34 -0
  1763. executorch/backends/xnnpack/_passes/__init__.py +109 -0
  1764. executorch/backends/xnnpack/_passes/channels_last_tagged_reshape_pass.py +561 -0
  1765. executorch/backends/xnnpack/_passes/conv1d_unsqueeze_pass.py +203 -0
  1766. executorch/backends/xnnpack/_passes/convert_to_linear.py +198 -0
  1767. executorch/backends/xnnpack/_passes/convert_to_sdpa.py +114 -0
  1768. executorch/backends/xnnpack/_passes/convert_to_upsample_bilinear2d.py +65 -0
  1769. executorch/backends/xnnpack/_passes/decompose_cat.py +105 -0
  1770. executorch/backends/xnnpack/_passes/fuse_activation_pass.py +87 -0
  1771. executorch/backends/xnnpack/_passes/fuse_batch_norm.py +261 -0
  1772. executorch/backends/xnnpack/_passes/prelu_reshape_pass.py +75 -0
  1773. executorch/backends/xnnpack/_passes/propagate_custom_meta_pass.py +44 -0
  1774. executorch/backends/xnnpack/_passes/remove_redundant_copy_pass.py +158 -0
  1775. executorch/backends/xnnpack/_passes/xnnpack_pass.py +22 -0
  1776. executorch/backends/xnnpack/operators/__init__.py +61 -0
  1777. executorch/backends/xnnpack/operators/node_visitor.py +758 -0
  1778. executorch/backends/xnnpack/operators/op_abs.py +52 -0
  1779. executorch/backends/xnnpack/operators/op_add.py +77 -0
  1780. executorch/backends/xnnpack/operators/op_addmm.py +70 -0
  1781. executorch/backends/xnnpack/operators/op_avg_pooling2d.py +75 -0
  1782. executorch/backends/xnnpack/operators/op_bmm.py +54 -0
  1783. executorch/backends/xnnpack/operators/op_cat.py +121 -0
  1784. executorch/backends/xnnpack/operators/op_ceiling.py +52 -0
  1785. executorch/backends/xnnpack/operators/op_clamp.py +64 -0
  1786. executorch/backends/xnnpack/operators/op_clone.py +60 -0
  1787. executorch/backends/xnnpack/operators/op_conv2d.py +186 -0
  1788. executorch/backends/xnnpack/operators/op_cos.py +52 -0
  1789. executorch/backends/xnnpack/operators/op_div.py +56 -0
  1790. executorch/backends/xnnpack/operators/op_dynamic_dequantize_ops.py +114 -0
  1791. executorch/backends/xnnpack/operators/op_dynamic_quantize_ops.py +179 -0
  1792. executorch/backends/xnnpack/operators/op_elu.py +56 -0
  1793. executorch/backends/xnnpack/operators/op_exp.py +52 -0
  1794. executorch/backends/xnnpack/operators/op_floor.py +48 -0
  1795. executorch/backends/xnnpack/operators/op_gelu.py +52 -0
  1796. executorch/backends/xnnpack/operators/op_hardswish.py +52 -0
  1797. executorch/backends/xnnpack/operators/op_hardtanh.py +67 -0
  1798. executorch/backends/xnnpack/operators/op_leaky_relu.py +65 -0
  1799. executorch/backends/xnnpack/operators/op_linear.py +110 -0
  1800. executorch/backends/xnnpack/operators/op_log.py +52 -0
  1801. executorch/backends/xnnpack/operators/op_matrix_multiplication.py +66 -0
  1802. executorch/backends/xnnpack/operators/op_max_dim.py +206 -0
  1803. executorch/backends/xnnpack/operators/op_max_pool2d.py +136 -0
  1804. executorch/backends/xnnpack/operators/op_maximum.py +55 -0
  1805. executorch/backends/xnnpack/operators/op_mean_dim.py +78 -0
  1806. executorch/backends/xnnpack/operators/op_minimum.py +54 -0
  1807. executorch/backends/xnnpack/operators/op_multiply.py +84 -0
  1808. executorch/backends/xnnpack/operators/op_negate.py +51 -0
  1809. executorch/backends/xnnpack/operators/op_permute.py +87 -0
  1810. executorch/backends/xnnpack/operators/op_prelu.py +56 -0
  1811. executorch/backends/xnnpack/operators/op_quant_dequant.py +196 -0
  1812. executorch/backends/xnnpack/operators/op_relu.py +61 -0
  1813. executorch/backends/xnnpack/operators/op_rsqrt.py +52 -0
  1814. executorch/backends/xnnpack/operators/op_sigmoid.py +48 -0
  1815. executorch/backends/xnnpack/operators/op_sin.py +52 -0
  1816. executorch/backends/xnnpack/operators/op_skip_ops.py +89 -0
  1817. executorch/backends/xnnpack/operators/op_slice_copy.py +100 -0
  1818. executorch/backends/xnnpack/operators/op_softmax.py +58 -0
  1819. executorch/backends/xnnpack/operators/op_square.py +52 -0
  1820. executorch/backends/xnnpack/operators/op_square_root.py +52 -0
  1821. executorch/backends/xnnpack/operators/op_squeeze.py +145 -0
  1822. executorch/backends/xnnpack/operators/op_static_constant_pad.py +145 -0
  1823. executorch/backends/xnnpack/operators/op_static_resize_bilinear_2d.py +61 -0
  1824. executorch/backends/xnnpack/operators/op_sub.py +77 -0
  1825. executorch/backends/xnnpack/operators/op_tanh.py +52 -0
  1826. executorch/backends/xnnpack/operators/op_to_copy.py +91 -0
  1827. executorch/backends/xnnpack/operators/op_view_copy.py +96 -0
  1828. executorch/backends/xnnpack/operators/quant_params.py +412 -0
  1829. executorch/backends/xnnpack/partition/config/__init__.py +127 -0
  1830. executorch/backends/xnnpack/partition/config/gemm_configs.py +575 -0
  1831. executorch/backends/xnnpack/partition/config/generic_node_configs.py +711 -0
  1832. executorch/backends/xnnpack/partition/config/node_configs.py +148 -0
  1833. executorch/backends/xnnpack/partition/config/quant_affine_configs.py +56 -0
  1834. executorch/backends/xnnpack/partition/config/xnnpack_config.py +252 -0
  1835. executorch/backends/xnnpack/partition/graphs/bilinear_2d.py +55 -0
  1836. executorch/backends/xnnpack/partition/graphs/sdpa.py +86 -0
  1837. executorch/backends/xnnpack/partition/xnnpack_partitioner.py +160 -0
  1838. executorch/backends/xnnpack/quantizer/xnnpack_quantizer.py +462 -0
  1839. executorch/backends/xnnpack/quantizer/xnnpack_quantizer_utils.py +1121 -0
  1840. executorch/backends/xnnpack/recipes/__init__.py +19 -0
  1841. executorch/backends/xnnpack/recipes/xnnpack_recipe_provider.py +200 -0
  1842. executorch/backends/xnnpack/recipes/xnnpack_recipe_types.py +37 -0
  1843. executorch/backends/xnnpack/serialization/runtime_schema.fbs +389 -0
  1844. executorch/backends/xnnpack/serialization/schema.fbs +381 -0
  1845. executorch/backends/xnnpack/serialization/xnnpack_graph_schema.py +542 -0
  1846. executorch/backends/xnnpack/serialization/xnnpack_graph_serialize.py +373 -0
  1847. executorch/backends/xnnpack/test/__init__.py +0 -0
  1848. executorch/backends/xnnpack/test/models/deeplab_v3.py +42 -0
  1849. executorch/backends/xnnpack/test/models/edsr.py +55 -0
  1850. executorch/backends/xnnpack/test/models/emformer_rnnt.py +98 -0
  1851. executorch/backends/xnnpack/test/models/inception_v3.py +84 -0
  1852. executorch/backends/xnnpack/test/models/inception_v4.py +62 -0
  1853. executorch/backends/xnnpack/test/models/llama2_et_example.py +49 -0
  1854. executorch/backends/xnnpack/test/models/mobilebert.py +55 -0
  1855. executorch/backends/xnnpack/test/models/mobilenet_v2.py +103 -0
  1856. executorch/backends/xnnpack/test/models/mobilenet_v3.py +84 -0
  1857. executorch/backends/xnnpack/test/models/resnet.py +90 -0
  1858. executorch/backends/xnnpack/test/models/torchvision_vit.py +105 -0
  1859. executorch/backends/xnnpack/test/models/very_big_model.py +45 -0
  1860. executorch/backends/xnnpack/test/models/w2l.py +60 -0
  1861. executorch/backends/xnnpack/test/ops/test_abs.py +87 -0
  1862. executorch/backends/xnnpack/test/ops/test_add.py +266 -0
  1863. executorch/backends/xnnpack/test/ops/test_avgpool2d.py +92 -0
  1864. executorch/backends/xnnpack/test/ops/test_bilinear2d.py +152 -0
  1865. executorch/backends/xnnpack/test/ops/test_bmm.py +49 -0
  1866. executorch/backends/xnnpack/test/ops/test_cat.py +254 -0
  1867. executorch/backends/xnnpack/test/ops/test_ceil.py +63 -0
  1868. executorch/backends/xnnpack/test/ops/test_check_quant_params.py +107 -0
  1869. executorch/backends/xnnpack/test/ops/test_clamp.py +79 -0
  1870. executorch/backends/xnnpack/test/ops/test_clone.py +137 -0
  1871. executorch/backends/xnnpack/test/ops/test_conv1d.py +190 -0
  1872. executorch/backends/xnnpack/test/ops/test_conv2d.py +747 -0
  1873. executorch/backends/xnnpack/test/ops/test_cos.py +87 -0
  1874. executorch/backends/xnnpack/test/ops/test_div.py +78 -0
  1875. executorch/backends/xnnpack/test/ops/test_elu.py +98 -0
  1876. executorch/backends/xnnpack/test/ops/test_exp.py +70 -0
  1877. executorch/backends/xnnpack/test/ops/test_floor.py +58 -0
  1878. executorch/backends/xnnpack/test/ops/test_gelu.py +71 -0
  1879. executorch/backends/xnnpack/test/ops/test_hardswish.py +70 -0
  1880. executorch/backends/xnnpack/test/ops/test_hardtanh.py +83 -0
  1881. executorch/backends/xnnpack/test/ops/test_leaky_relu.py +131 -0
  1882. executorch/backends/xnnpack/test/ops/test_linear.py +1059 -0
  1883. executorch/backends/xnnpack/test/ops/test_log.py +45 -0
  1884. executorch/backends/xnnpack/test/ops/test_lstm.py +67 -0
  1885. executorch/backends/xnnpack/test/ops/test_max_dim.py +66 -0
  1886. executorch/backends/xnnpack/test/ops/test_maximum.py +66 -0
  1887. executorch/backends/xnnpack/test/ops/test_maxpool2d.py +221 -0
  1888. executorch/backends/xnnpack/test/ops/test_mean_dim.py +98 -0
  1889. executorch/backends/xnnpack/test/ops/test_minimum.py +49 -0
  1890. executorch/backends/xnnpack/test/ops/test_multiply.py +149 -0
  1891. executorch/backends/xnnpack/test/ops/test_negate.py +58 -0
  1892. executorch/backends/xnnpack/test/ops/test_permute.py +134 -0
  1893. executorch/backends/xnnpack/test/ops/test_pow.py +62 -0
  1894. executorch/backends/xnnpack/test/ops/test_prelu.py +50 -0
  1895. executorch/backends/xnnpack/test/ops/test_quantize_per_tensor.py +70 -0
  1896. executorch/backends/xnnpack/test/ops/test_relu.py +38 -0
  1897. executorch/backends/xnnpack/test/ops/test_rsqrt.py +45 -0
  1898. executorch/backends/xnnpack/test/ops/test_sigmoid.py +45 -0
  1899. executorch/backends/xnnpack/test/ops/test_sin.py +87 -0
  1900. executorch/backends/xnnpack/test/ops/test_slice_copy.py +170 -0
  1901. executorch/backends/xnnpack/test/ops/test_softmax.py +73 -0
  1902. executorch/backends/xnnpack/test/ops/test_sqrt.py +45 -0
  1903. executorch/backends/xnnpack/test/ops/test_square.py +48 -0
  1904. executorch/backends/xnnpack/test/ops/test_static_constant_pad.py +294 -0
  1905. executorch/backends/xnnpack/test/ops/test_sub.py +178 -0
  1906. executorch/backends/xnnpack/test/ops/test_tanh.py +43 -0
  1907. executorch/backends/xnnpack/test/ops/test_to_copy.py +113 -0
  1908. executorch/backends/xnnpack/test/passes/__init__.py +0 -0
  1909. executorch/backends/xnnpack/test/passes/test_activation_fusion.py +232 -0
  1910. executorch/backends/xnnpack/test/passes/test_batch_norm_fusion.py +179 -0
  1911. executorch/backends/xnnpack/test/passes/test_channels_last_tagged_reshape.py +635 -0
  1912. executorch/backends/xnnpack/test/passes/test_convert_to_linear.py +43 -0
  1913. executorch/backends/xnnpack/test/passes/test_decompose_cat_pass.py +112 -0
  1914. executorch/backends/xnnpack/test/passes/test_propagate_custom_meta_pass.py +185 -0
  1915. executorch/backends/xnnpack/test/passes/test_remove_get_item_pass.py +99 -0
  1916. executorch/backends/xnnpack/test/passes/test_remove_redundant_copy_pass.py +182 -0
  1917. executorch/backends/xnnpack/test/quantizer/test_pt2e_quantization.py +818 -0
  1918. executorch/backends/xnnpack/test/quantizer/test_representation.py +308 -0
  1919. executorch/backends/xnnpack/test/quantizer/test_xnnpack_quantizer.py +1217 -0
  1920. executorch/backends/xnnpack/test/recipes/test_xnnpack_recipes.py +320 -0
  1921. executorch/backends/xnnpack/test/serialization/test_serialization.py +50 -0
  1922. executorch/backends/xnnpack/test/serialization/test_xnnheader.py +104 -0
  1923. executorch/backends/xnnpack/test/test_xnnpack_partitioner.py +163 -0
  1924. executorch/backends/xnnpack/test/test_xnnpack_utils.py +532 -0
  1925. executorch/backends/xnnpack/test/test_xnnpack_utils_classes.py +44 -0
  1926. executorch/backends/xnnpack/test/tester/__init__.py +29 -0
  1927. executorch/backends/xnnpack/test/tester/tester.py +130 -0
  1928. executorch/backends/xnnpack/third-party/FP16/configure.py +65 -0
  1929. executorch/backends/xnnpack/third-party/FP16/confu.yaml +8 -0
  1930. executorch/backends/xnnpack/third-party/FP16/include/fp16/__init__.py +0 -0
  1931. executorch/backends/xnnpack/third-party/FP16/include/fp16/avx.py +50 -0
  1932. executorch/backends/xnnpack/third-party/FP16/include/fp16/avx2.py +53 -0
  1933. executorch/backends/xnnpack/third-party/FP16/test/peachpy/stubs.py +38 -0
  1934. executorch/backends/xnnpack/third-party/FXdiv/configure.py +30 -0
  1935. executorch/backends/xnnpack/third-party/FXdiv/confu.yaml +6 -0
  1936. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/aarch32_template.py +361 -0
  1937. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/aarch64_template.py +347 -0
  1938. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/arm_template.py +78 -0
  1939. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512bf16_template.py +81 -0
  1940. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512f_template.py +390 -0
  1941. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512vnni_template.py +703 -0
  1942. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/base_architecture.py +276 -0
  1943. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/fma3_template.py +260 -0
  1944. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate.py +71 -0
  1945. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_bf16_f32_gemm_microkernels.py +39 -0
  1946. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_f32_gemm_microkernels.py +104 -0
  1947. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_gemm_microkernels_main.py +29 -0
  1948. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qd8_f32_qc4w_gemm_microkernels.py +69 -0
  1949. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qd8_f32_qc8w_gemm_microkernels.py +93 -0
  1950. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qs8_qc4w_gemm_microkernels.py +42 -0
  1951. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qs8_qc8w_gemm_microkernels.py +43 -0
  1952. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neondot_template.py +463 -0
  1953. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neonfma_template.py +248 -0
  1954. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neonmlal_aarch32_template.py +394 -0
  1955. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/x64_template.py +512 -0
  1956. executorch/backends/xnnpack/third-party/XNNPACK/scripts/check_config.py +152 -0
  1957. executorch/backends/xnnpack/third-party/XNNPACK/scripts/check_files_changed.py +53 -0
  1958. executorch/backends/xnnpack/third-party/XNNPACK/scripts/generate-build-identifier.py +106 -0
  1959. executorch/backends/xnnpack/third-party/XNNPACK/scripts/sort-filenames.py +134 -0
  1960. executorch/backends/xnnpack/third-party/XNNPACK/test/bf16-f32-gemm-minmax.yaml +121 -0
  1961. executorch/backends/xnnpack/third-party/XNNPACK/test/bf16-gemm-minmax.yaml +106 -0
  1962. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-conv-hwc2chw.yaml +8 -0
  1963. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-dwconv2d-chw.yaml +88 -0
  1964. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-f32acc-gemm-minmax.yaml +42 -0
  1965. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-f32acc-igemm-minmax.yaml +42 -0
  1966. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-gemm-minmax.yaml +219 -0
  1967. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-ibilinear-chw.yaml +9 -0
  1968. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-ibilinear.yaml +12 -0
  1969. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-igemm-minmax.yaml +165 -0
  1970. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-raddstoreexpminusmax.yaml +52 -0
  1971. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-argmaxpool.yaml +15 -0
  1972. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-conv-hwc2chw.yaml +32 -0
  1973. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-dwconv2d-chw.yaml +712 -0
  1974. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm-minmax.yaml +1572 -0
  1975. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm-relu.yaml +123 -0
  1976. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm.yaml +127 -0
  1977. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ibilinear-chw.yaml +22 -0
  1978. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ibilinear.yaml +23 -0
  1979. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm-minmax.yaml +1001 -0
  1980. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm-relu.yaml +123 -0
  1981. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm.yaml +123 -0
  1982. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ppmm-minmax.yaml +122 -0
  1983. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc4w-gemm-minmax.yaml +277 -0
  1984. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm-minmax.yaml +757 -0
  1985. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm-relu.yaml +67 -0
  1986. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm.yaml +115 -0
  1987. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-raddexpminusmax.yaml +38 -0
  1988. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-raddstoreexpminusmax.yaml +88 -0
  1989. executorch/backends/xnnpack/third-party/XNNPACK/test/pf16-gemm-minmax.yaml +20 -0
  1990. executorch/backends/xnnpack/third-party/XNNPACK/test/pf32-gemm-minmax.yaml +34 -0
  1991. executorch/backends/xnnpack/third-party/XNNPACK/test/pqs8-qc8w-gemm-minmax.yaml +20 -0
  1992. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qb4w-gemm-minmax.yaml +260 -0
  1993. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc4w-gemm-minmax.yaml +678 -0
  1994. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc8w-gemm-minmax.yaml +412 -0
  1995. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc8w-igemm-minmax.yaml +360 -0
  1996. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qb4w-gemm-minmax.yaml +537 -0
  1997. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc4w-gemm-minmax.yaml +2176 -0
  1998. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc8w-gemm-minmax.yaml +1564 -0
  1999. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc8w-igemm-minmax.yaml +1035 -0
  2000. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qb4w-gemm-minmax.yaml +54 -0
  2001. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qc4w-gemm-minmax.yaml +66 -0
  2002. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qc8w-gemm-minmax.yaml +42 -0
  2003. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc4w-gemm-minmax-fp32.yaml +345 -0
  2004. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc8w-gemm-minmax-fp32.yaml +2029 -0
  2005. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc8w-igemm-minmax-fp32.yaml +1831 -0
  2006. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-gemm-minmax-fp32.yaml +584 -0
  2007. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-gemm-minmax-rndnu.yaml +148 -0
  2008. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-igemm-minmax-fp32.yaml +581 -0
  2009. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-igemm-minmax-rndnu.yaml +110 -0
  2010. executorch/backends/xnnpack/third-party/XNNPACK/test/s8-ibilinear.yaml +22 -0
  2011. executorch/backends/xnnpack/third-party/XNNPACK/test/u8-ibilinear.yaml +22 -0
  2012. executorch/backends/xnnpack/third-party/XNNPACK/test/x8-lut.yaml +47 -0
  2013. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-argmaxpool-test.py +692 -0
  2014. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-conv-hwc-test.py +543 -0
  2015. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-conv-hwc2chw-test.py +591 -0
  2016. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-dwconv-test.py +339 -0
  2017. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-dwconv2d-chw-test.py +463 -0
  2018. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-gemm-test.py +1150 -0
  2019. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-ibilinear-chw-test.py +236 -0
  2020. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-ibilinear-test.py +237 -0
  2021. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-lut-test.py +164 -0
  2022. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-raddexpminusmax-test.py +159 -0
  2023. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-raddstoreexpminusmax-test.py +210 -0
  2024. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-reduce-discontiguous-test.py +495 -0
  2025. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-reduce-test.py +289 -0
  2026. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-spmm-test.py +554 -0
  2027. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-vbinary-test.py +179 -0
  2028. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-vunary-test.py +280 -0
  2029. executorch/backends/xnnpack/third-party/XNNPACK/tools/parse-microkernel-bench.py +189 -0
  2030. executorch/backends/xnnpack/third-party/XNNPACK/tools/primes.py +1015 -0
  2031. executorch/backends/xnnpack/third-party/XNNPACK/tools/update-microkernels.py +567 -0
  2032. executorch/backends/xnnpack/third-party/XNNPACK/tools/xngen.py +159 -0
  2033. executorch/backends/xnnpack/third-party/XNNPACK/tools/xnncommon.py +275 -0
  2034. executorch/backends/xnnpack/third-party/cpuinfo/configure.py +110 -0
  2035. executorch/backends/xnnpack/third-party/cpuinfo/confu.yaml +9 -0
  2036. executorch/backends/xnnpack/third-party/cpuinfo/deps/clog/configure.py +26 -0
  2037. executorch/backends/xnnpack/third-party/cpuinfo/deps/clog/confu.yaml +5 -0
  2038. executorch/backends/xnnpack/third-party/cpuinfo/scripts/android-device-dump.py +400 -0
  2039. executorch/backends/xnnpack/third-party/cpuinfo/scripts/arm-linux-filesystem-dump.py +107 -0
  2040. executorch/backends/xnnpack/third-party/cpuinfo/scripts/parse-x86-cpuid-dump.py +47 -0
  2041. executorch/backends/xnnpack/third-party/generate-cpuinfo-wrappers.py +93 -0
  2042. executorch/backends/xnnpack/third-party/pthreadpool/configure.py +38 -0
  2043. executorch/backends/xnnpack/third-party/pthreadpool/confu.yaml +8 -0
  2044. executorch/backends/xnnpack/utils/configs.py +49 -0
  2045. executorch/backends/xnnpack/utils/quant_utils.py +301 -0
  2046. executorch/backends/xnnpack/utils/utils.py +221 -0
  2047. executorch/backends/xnnpack/utils/xnnpack_constants.py +37 -0
  2048. executorch/backends/xnnpack/xnnpack_preprocess.py +194 -0
  2049. executorch/codegen/__init__.py +0 -0
  2050. executorch/codegen/api/__init__.py +0 -0
  2051. executorch/codegen/api/custom_ops.py +151 -0
  2052. executorch/codegen/api/et_cpp.py +368 -0
  2053. executorch/codegen/api/types/__init__.py +5 -0
  2054. executorch/codegen/api/types/signatures.py +77 -0
  2055. executorch/codegen/api/types/types.py +77 -0
  2056. executorch/codegen/api/unboxing.py +232 -0
  2057. executorch/codegen/gen.py +1049 -0
  2058. executorch/codegen/model.py +220 -0
  2059. executorch/codegen/parse.py +158 -0
  2060. executorch/codegen/test/test_executorch_custom_ops.py +153 -0
  2061. executorch/codegen/test/test_executorch_gen.py +698 -0
  2062. executorch/codegen/test/test_executorch_signatures.py +65 -0
  2063. executorch/codegen/test/test_executorch_types.py +121 -0
  2064. executorch/codegen/test/test_executorch_unboxing.py +165 -0
  2065. executorch/codegen/test/test_selective_build.py +51 -0
  2066. executorch/codegen/tools/__init__.py +0 -0
  2067. executorch/codegen/tools/combine_prim_ops_headers.py +164 -0
  2068. executorch/codegen/tools/gen_all_oplist.py +252 -0
  2069. executorch/codegen/tools/gen_oplist.py +369 -0
  2070. executorch/codegen/tools/gen_ops_def.py +96 -0
  2071. executorch/codegen/tools/gen_selected_op_variants.py +133 -0
  2072. executorch/codegen/tools/gen_selected_prim_ops.py +96 -0
  2073. executorch/codegen/tools/merge_yaml.py +125 -0
  2074. executorch/codegen/tools/selective_build.cp313-win_amd64.pyd +0 -0
  2075. executorch/codegen/tools/selective_build.pyi +23 -0
  2076. executorch/codegen/tools/test/test_gen_all_oplist.py +122 -0
  2077. executorch/codegen/tools/test/test_gen_oplist.py +263 -0
  2078. executorch/codegen/tools/test/test_gen_oplist_real_model.py +50 -0
  2079. executorch/codegen/tools/test/test_gen_selected_op_variants.py +204 -0
  2080. executorch/codegen/tools/test/test_tools_selective_build.py +129 -0
  2081. executorch/codegen/tools/yaml_util.py +15 -0
  2082. executorch/data/bin/__init__.py +44 -0
  2083. executorch/data/bin/flatc.exe +0 -0
  2084. executorch/data/lib/aoti_cuda_shims.lib +0 -0
  2085. executorch/devtools/__init__.py +19 -0
  2086. executorch/devtools/backend_debug/__init__.py +13 -0
  2087. executorch/devtools/backend_debug/delegation_info.py +183 -0
  2088. executorch/devtools/backend_debug/tests/test_delegation_info.py +79 -0
  2089. executorch/devtools/bundled_program/config.py +120 -0
  2090. executorch/devtools/bundled_program/core.py +395 -0
  2091. executorch/devtools/bundled_program/schema/__init__.py +9 -0
  2092. executorch/devtools/bundled_program/schema/bundled_program_schema.fbs +97 -0
  2093. executorch/devtools/bundled_program/schema/bundled_program_schema.py +99 -0
  2094. executorch/devtools/bundled_program/schema/scalar_type.fbs +43 -0
  2095. executorch/devtools/bundled_program/schema/test/test_schema.py +27 -0
  2096. executorch/devtools/bundled_program/serialize/__init__.py +122 -0
  2097. executorch/devtools/bundled_program/serialize/bundled_program_schema.fbs +97 -0
  2098. executorch/devtools/bundled_program/serialize/scalar_type.fbs +43 -0
  2099. executorch/devtools/bundled_program/serialize/test/test_serialize.py +37 -0
  2100. executorch/devtools/bundled_program/test/test_bundle_data.py +168 -0
  2101. executorch/devtools/bundled_program/test/test_config.py +131 -0
  2102. executorch/devtools/bundled_program/test/test_end2end.py +64 -0
  2103. executorch/devtools/bundled_program/util/test_util.py +289 -0
  2104. executorch/devtools/bundled_program/version.py +13 -0
  2105. executorch/devtools/debug_format/base_schema.py +61 -0
  2106. executorch/devtools/debug_format/et_schema.py +432 -0
  2107. executorch/devtools/etdump/etdump_schema_flatcc.fbs +181 -0
  2108. executorch/devtools/etdump/scalar_type.fbs +43 -0
  2109. executorch/devtools/etdump/schema_flatcc.py +154 -0
  2110. executorch/devtools/etdump/serialize.py +108 -0
  2111. executorch/devtools/etdump/tests/serialize_test.py +142 -0
  2112. executorch/devtools/etrecord/__init__.py +13 -0
  2113. executorch/devtools/etrecord/_etrecord.py +776 -0
  2114. executorch/devtools/etrecord/tests/etrecord_test.py +1748 -0
  2115. executorch/devtools/inspector/__init__.py +24 -0
  2116. executorch/devtools/inspector/_inspector.py +1495 -0
  2117. executorch/devtools/inspector/_inspector_utils.py +1286 -0
  2118. executorch/devtools/inspector/_intermediate_output_capturer.py +62 -0
  2119. executorch/devtools/inspector/inspector_cli.py +76 -0
  2120. executorch/devtools/inspector/numerical_comparator/__init__.py +25 -0
  2121. executorch/devtools/inspector/numerical_comparator/l1_numerical_comparator.py +27 -0
  2122. executorch/devtools/inspector/numerical_comparator/mse_numerical_comparator.py +29 -0
  2123. executorch/devtools/inspector/numerical_comparator/numerical_comparator_base.py +26 -0
  2124. executorch/devtools/inspector/numerical_comparator/snr_numerical_comparator.py +39 -0
  2125. executorch/devtools/inspector/tests/event_blocks_test.py +657 -0
  2126. executorch/devtools/inspector/tests/inspector_test.py +1263 -0
  2127. executorch/devtools/inspector/tests/inspector_test_utils.py +189 -0
  2128. executorch/devtools/inspector/tests/inspector_utils_test.py +896 -0
  2129. executorch/devtools/inspector/tests/intermediate_output_capturer_test.py +101 -0
  2130. executorch/devtools/inspector/tests/l1_comparator_test.py +49 -0
  2131. executorch/devtools/inspector/tests/mse_comparator_test.py +49 -0
  2132. executorch/devtools/inspector/tests/snr_comparator_test.py +52 -0
  2133. executorch/devtools/scripts/generate_profiling_csv.py +62 -0
  2134. executorch/devtools/size_analysis_tool/size_analysis_tool.py +180 -0
  2135. executorch/devtools/size_analysis_tool/size_analysis_tool_test.py +101 -0
  2136. executorch/devtools/visualization/__init__.py +13 -0
  2137. executorch/devtools/visualization/visualization_utils.py +301 -0
  2138. executorch/devtools/visualization/visualization_utils_test.py +177 -0
  2139. executorch/examples/apple/coreml/llama/export.py +224 -0
  2140. executorch/examples/apple/coreml/llama/export_static_llm_coreml.py +504 -0
  2141. executorch/examples/apple/coreml/llama/llama_transformer.py +958 -0
  2142. executorch/examples/apple/coreml/llama/run.py +206 -0
  2143. executorch/examples/apple/coreml/llama/run_lookahead.py +284 -0
  2144. executorch/examples/apple/coreml/llama/run_static_llm.py +293 -0
  2145. executorch/examples/apple/coreml/llama/test.py +48 -0
  2146. executorch/examples/apple/coreml/llama/utils.py +116 -0
  2147. executorch/examples/apple/coreml/scripts/debugger_cli.py +181 -0
  2148. executorch/examples/apple/coreml/scripts/export.py +280 -0
  2149. executorch/examples/apple/coreml/scripts/extract_coreml_models.py +77 -0
  2150. executorch/examples/apple/coreml/scripts/inspector_cli.py +64 -0
  2151. executorch/examples/apple/coreml/scripts/inspector_utils.py +425 -0
  2152. executorch/examples/apple/mps/scripts/bench_utils.py +108 -0
  2153. executorch/examples/apple/mps/scripts/mps_example.py +230 -0
  2154. executorch/examples/cuda/scripts/__init__.py +0 -0
  2155. executorch/examples/cuda/scripts/export.py +107 -0
  2156. executorch/examples/llm_pte_finetuning/__init__.py +19 -0
  2157. executorch/examples/llm_pte_finetuning/llama3_config.yaml +96 -0
  2158. executorch/examples/llm_pte_finetuning/model_exporter.py +87 -0
  2159. executorch/examples/llm_pte_finetuning/model_loading_lib.py +143 -0
  2160. executorch/examples/llm_pte_finetuning/phi3_alpaca_code_config.yaml +44 -0
  2161. executorch/examples/llm_pte_finetuning/phi3_config.yaml +42 -0
  2162. executorch/examples/llm_pte_finetuning/qwen_05b_config.yaml +43 -0
  2163. executorch/examples/llm_pte_finetuning/runner.py +135 -0
  2164. executorch/examples/llm_pte_finetuning/training_lib.py +157 -0
  2165. executorch/examples/models/__init__.py +98 -0
  2166. executorch/examples/models/checkpoint.py +114 -0
  2167. executorch/examples/models/codegen/__init__.py +16 -0
  2168. executorch/examples/models/codegen/convert_weight.py +93 -0
  2169. executorch/examples/models/deeplab_v3/__init__.py +9 -0
  2170. executorch/examples/models/deeplab_v3/model.py +50 -0
  2171. executorch/examples/models/deepseek-r1-distill-llama-8B/config/deepseek_xnnpack_q8da4w.yaml +16 -0
  2172. executorch/examples/models/deit_tiny/__init__.py +8 -0
  2173. executorch/examples/models/deit_tiny/model.py +36 -0
  2174. executorch/examples/models/edsr/__init__.py +11 -0
  2175. executorch/examples/models/edsr/model.py +28 -0
  2176. executorch/examples/models/efficient_sam/__init__.py +11 -0
  2177. executorch/examples/models/efficient_sam/efficient_sam_core/build_efficient_sam.py +25 -0
  2178. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam.py +316 -0
  2179. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam_decoder.py +373 -0
  2180. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam_encoder.py +261 -0
  2181. executorch/examples/models/efficient_sam/efficient_sam_core/mlp.py +31 -0
  2182. executorch/examples/models/efficient_sam/efficient_sam_core/two_way_transformer.py +268 -0
  2183. executorch/examples/models/efficient_sam/model.py +34 -0
  2184. executorch/examples/models/emformer_rnnt/__init__.py +17 -0
  2185. executorch/examples/models/emformer_rnnt/model.py +132 -0
  2186. executorch/examples/models/gemma/__init__.py +16 -0
  2187. executorch/examples/models/gemma/convert_weights.py +104 -0
  2188. executorch/examples/models/gemma2/__init__.py +19 -0
  2189. executorch/examples/models/gemma2/convert_weights.py +113 -0
  2190. executorch/examples/models/gemma3/__init__.py +16 -0
  2191. executorch/examples/models/gemma3/convert_weights.py +110 -0
  2192. executorch/examples/models/glm/__init__.py +16 -0
  2193. executorch/examples/models/glm/convert_weights.py +79 -0
  2194. executorch/examples/models/granite/__init__.py +16 -0
  2195. executorch/examples/models/granite/convert_weights.py +106 -0
  2196. executorch/examples/models/inception_v3/__init__.py +11 -0
  2197. executorch/examples/models/inception_v3/model.py +28 -0
  2198. executorch/examples/models/inception_v4/__init__.py +11 -0
  2199. executorch/examples/models/inception_v4/model.py +26 -0
  2200. executorch/examples/models/internvl3/__init__.py +16 -0
  2201. executorch/examples/models/internvl3/convert_weights.py +82 -0
  2202. executorch/examples/models/lfm2/__init__.py +10 -0
  2203. executorch/examples/models/lfm2/config/lfm2_xnnpack_fp32.yaml +12 -0
  2204. executorch/examples/models/lfm2/config/lfm2_xnnpack_q8da4w.yaml +15 -0
  2205. executorch/examples/models/lfm2/convert_weights.py +74 -0
  2206. executorch/examples/models/lfm2/short_conv.py +112 -0
  2207. executorch/examples/models/llama/__init__.py +11 -0
  2208. executorch/examples/models/llama/attention.py +536 -0
  2209. executorch/examples/models/llama/config/llama_bf16.yaml +7 -0
  2210. executorch/examples/models/llama/config/llama_q8da4w.yaml +11 -0
  2211. executorch/examples/models/llama/config/llama_xnnpack.yaml +17 -0
  2212. executorch/examples/models/llama/config/llama_xnnpack_qat.yaml +23 -0
  2213. executorch/examples/models/llama/config/llama_xnnpack_spinquant.yaml +22 -0
  2214. executorch/examples/models/llama/config/test_llm_config.py +104 -0
  2215. executorch/examples/models/llama/convert_weights.py +61 -0
  2216. executorch/examples/models/llama/eval_llama.py +38 -0
  2217. executorch/examples/models/llama/eval_llama_lib.py +395 -0
  2218. executorch/examples/models/llama/evaluate/__init__.py +11 -0
  2219. executorch/examples/models/llama/evaluate/eager_eval.py +87 -0
  2220. executorch/examples/models/llama/experimental/__init__.py +0 -0
  2221. executorch/examples/models/llama/experimental/generate.py +190 -0
  2222. executorch/examples/models/llama/experimental/load_gguf_q4_0.py +184 -0
  2223. executorch/examples/models/llama/experimental/subclass.py +413 -0
  2224. executorch/examples/models/llama/experimental/test_subclass.py +28 -0
  2225. executorch/examples/models/llama/export_llama.py +56 -0
  2226. executorch/examples/models/llama/export_llama_lib.py +1560 -0
  2227. executorch/examples/models/llama/fairseq2.py +63 -0
  2228. executorch/examples/models/llama/feed_forward.py +68 -0
  2229. executorch/examples/models/llama/hf_download.py +50 -0
  2230. executorch/examples/models/llama/install_requirement_helper.py +29 -0
  2231. executorch/examples/models/llama/llama_transformer.py +289 -0
  2232. executorch/examples/models/llama/lora.py +48 -0
  2233. executorch/examples/models/llama/model.py +378 -0
  2234. executorch/examples/models/llama/model_args.py +173 -0
  2235. executorch/examples/models/llama/norm.py +55 -0
  2236. executorch/examples/models/llama/rope.py +327 -0
  2237. executorch/examples/models/llama/runner/eager.py +142 -0
  2238. executorch/examples/models/llama/runner/generation.py +234 -0
  2239. executorch/examples/models/llama/runner/native.py +145 -0
  2240. executorch/examples/models/llama/source_transformation/__init__.py +0 -0
  2241. executorch/examples/models/llama/source_transformation/apply_spin_quant_r1_r2.py +202 -0
  2242. executorch/examples/models/llama/source_transformation/attention.py +223 -0
  2243. executorch/examples/models/llama/source_transformation/attention_sink.py +312 -0
  2244. executorch/examples/models/llama/source_transformation/custom_kv_cache.py +598 -0
  2245. executorch/examples/models/llama/source_transformation/lora.py +153 -0
  2246. executorch/examples/models/llama/source_transformation/pre_quantization.py +204 -0
  2247. executorch/examples/models/llama/source_transformation/prune_vocab.py +119 -0
  2248. executorch/examples/models/llama/source_transformation/quantize.py +910 -0
  2249. executorch/examples/models/llama/source_transformation/rms_norm.py +23 -0
  2250. executorch/examples/models/llama/source_transformation/rope.py +39 -0
  2251. executorch/examples/models/llama/source_transformation/sdpa.py +550 -0
  2252. executorch/examples/models/llama/source_transformation/spin_quant.py +88 -0
  2253. executorch/examples/models/llama/source_transformation/test_attention_sink.py +514 -0
  2254. executorch/examples/models/llama/source_transformation/test_quantized_kv_cache.py +126 -0
  2255. executorch/examples/models/llama/source_transformation/test_quantized_sdpa.py +167 -0
  2256. executorch/examples/models/llama/source_transformation/test_sdpa_with_quantized_kv_cache.py +97 -0
  2257. executorch/examples/models/llama/static_attention.py +1257 -0
  2258. executorch/examples/models/llama/tests/test_export_llama_lib.py +50 -0
  2259. executorch/examples/models/llama/tests/test_pre_quantization_transforms.py +188 -0
  2260. executorch/examples/models/llama/tests/test_replace_kv_cache.py +158 -0
  2261. executorch/examples/models/llama/tests/test_ring_attention.py +359 -0
  2262. executorch/examples/models/llama/tests/test_ring_kv_cache.py +492 -0
  2263. executorch/examples/models/llama/tests/test_simple_sdpa.py +66 -0
  2264. executorch/examples/models/llama/tests/test_static_attention.py +363 -0
  2265. executorch/examples/models/llama/tokenizer/tiktoken.py +224 -0
  2266. executorch/examples/models/llama3_2_vision/__init__.py +14 -0
  2267. executorch/examples/models/llama3_2_vision/preprocess/__init__.py +0 -0
  2268. executorch/examples/models/llama3_2_vision/preprocess/export_preprocess.py +45 -0
  2269. executorch/examples/models/llama3_2_vision/preprocess/model.py +72 -0
  2270. executorch/examples/models/llama3_2_vision/preprocess/test_preprocess.py +338 -0
  2271. executorch/examples/models/llama3_2_vision/runner/eager.py +59 -0
  2272. executorch/examples/models/llama3_2_vision/runner/exported.py +94 -0
  2273. executorch/examples/models/llama3_2_vision/runner/generation.py +102 -0
  2274. executorch/examples/models/llama3_2_vision/runner/native.py +143 -0
  2275. executorch/examples/models/llama3_2_vision/text_decoder/model.py +199 -0
  2276. executorch/examples/models/llama3_2_vision/text_decoder/test/__init__.py +0 -0
  2277. executorch/examples/models/llama3_2_vision/text_decoder/test/test_text_decoder.py +89 -0
  2278. executorch/examples/models/llama3_2_vision/vision_encoder/__init__.py +12 -0
  2279. executorch/examples/models/llama3_2_vision/vision_encoder/model.py +96 -0
  2280. executorch/examples/models/llama3_2_vision/vision_encoder/test/__init__.py +0 -0
  2281. executorch/examples/models/llama3_2_vision/vision_encoder/test/test_vision_encoder.py +46 -0
  2282. executorch/examples/models/llava/__init__.py +11 -0
  2283. executorch/examples/models/llava/export_llava.py +365 -0
  2284. executorch/examples/models/llava/model.py +415 -0
  2285. executorch/examples/models/llava/test/test_llava.py +151 -0
  2286. executorch/examples/models/llava/test/test_pte.py +106 -0
  2287. executorch/examples/models/lstm/__init__.py +12 -0
  2288. executorch/examples/models/lstm/model.py +31 -0
  2289. executorch/examples/models/mobilebert/__init__.py +11 -0
  2290. executorch/examples/models/mobilebert/model.py +31 -0
  2291. executorch/examples/models/mobilenet_v2/__init__.py +12 -0
  2292. executorch/examples/models/mobilenet_v2/model.py +73 -0
  2293. executorch/examples/models/mobilenet_v3/__init__.py +11 -0
  2294. executorch/examples/models/mobilenet_v3/model.py +30 -0
  2295. executorch/examples/models/model_base.py +47 -0
  2296. executorch/examples/models/model_factory.py +60 -0
  2297. executorch/examples/models/moshi/mimi/test_mimi.py +245 -0
  2298. executorch/examples/models/parakeet/export_parakeet_tdt.py +591 -0
  2299. executorch/examples/models/phi-3-mini/__init__.py +11 -0
  2300. executorch/examples/models/phi-3-mini-lora/export_model.py +144 -0
  2301. executorch/examples/models/phi_4_mini/__init__.py +16 -0
  2302. executorch/examples/models/phi_4_mini/config/phi_4_mini_xnnpack.yaml +12 -0
  2303. executorch/examples/models/phi_4_mini/convert_weights.py +151 -0
  2304. executorch/examples/models/qwen2_5/__init__.py +16 -0
  2305. executorch/examples/models/qwen2_5/config/qwen2_5_xnnpack_q8da4w.yaml +14 -0
  2306. executorch/examples/models/qwen2_5/convert_weights.py +90 -0
  2307. executorch/examples/models/qwen3/__init__.py +16 -0
  2308. executorch/examples/models/qwen3/config/qwen3_xnnpack.yaml +18 -0
  2309. executorch/examples/models/qwen3/config/qwen3_xnnpack_q8da4w.yaml +20 -0
  2310. executorch/examples/models/qwen3/convert_weights.py +140 -0
  2311. executorch/examples/models/resnet/__init__.py +12 -0
  2312. executorch/examples/models/resnet/model.py +48 -0
  2313. executorch/examples/models/smollm2/__init__.py +16 -0
  2314. executorch/examples/models/smollm2/convert_weights.py +84 -0
  2315. executorch/examples/models/smollm3/__init__.py +16 -0
  2316. executorch/examples/models/smollm3/convert_weights.py +112 -0
  2317. executorch/examples/models/smolvlm/__init__.py +16 -0
  2318. executorch/examples/models/smolvlm/convert_weights.py +77 -0
  2319. executorch/examples/models/stable_diffusion/__init__.py +9 -0
  2320. executorch/examples/models/stable_diffusion/model.py +193 -0
  2321. executorch/examples/models/test/__init__.py +0 -0
  2322. executorch/examples/models/test/test_export.py +161 -0
  2323. executorch/examples/models/torchvision_vit/__init__.py +11 -0
  2324. executorch/examples/models/torchvision_vit/model.py +28 -0
  2325. executorch/examples/models/toy_model/__init__.py +25 -0
  2326. executorch/examples/models/toy_model/model.py +143 -0
  2327. executorch/examples/models/wav2letter/__init__.py +11 -0
  2328. executorch/examples/models/wav2letter/model.py +29 -0
  2329. executorch/examples/models/yolo12/export_and_validate.py +397 -0
  2330. executorch/examples/nxp/experimental/cifar_net/cifar_net.py +270 -0
  2331. executorch/examples/xnnpack/__init__.py +51 -0
  2332. executorch/examples/xnnpack/aot_compiler.py +148 -0
  2333. executorch/examples/xnnpack/quantization/example.py +205 -0
  2334. executorch/examples/xnnpack/quantization/utils.py +42 -0
  2335. executorch/exir/__init__.py +57 -0
  2336. executorch/exir/_serialize/__init__.py +20 -0
  2337. executorch/exir/_serialize/_cord.py +49 -0
  2338. executorch/exir/_serialize/_dataclass.py +145 -0
  2339. executorch/exir/_serialize/_flatbuffer.py +357 -0
  2340. executorch/exir/_serialize/_named_data_store.py +222 -0
  2341. executorch/exir/_serialize/_program.py +767 -0
  2342. executorch/exir/_serialize/_serialize.py +123 -0
  2343. executorch/exir/_serialize/data_serializer.py +80 -0
  2344. executorch/exir/_serialize/padding.py +39 -0
  2345. executorch/exir/_serialize/program.fbs +489 -0
  2346. executorch/exir/_serialize/scalar_type.fbs +43 -0
  2347. executorch/exir/_serialize/test/__init__.py +0 -0
  2348. executorch/exir/_serialize/test/test_cord.py +63 -0
  2349. executorch/exir/_serialize/test/test_flatbuffer.py +331 -0
  2350. executorch/exir/_serialize/test/test_named_data_store.py +212 -0
  2351. executorch/exir/_serialize/test/test_program.py +1265 -0
  2352. executorch/exir/_serialize/test/test_serialize.py +289 -0
  2353. executorch/exir/_warnings.py +44 -0
  2354. executorch/exir/backend/backend_api.py +794 -0
  2355. executorch/exir/backend/backend_details.py +150 -0
  2356. executorch/exir/backend/canonical_partitioners/all_node_partitioner.py +55 -0
  2357. executorch/exir/backend/canonical_partitioners/config_partitioner.py +262 -0
  2358. executorch/exir/backend/canonical_partitioners/duplicate_constant_node_pass.py +156 -0
  2359. executorch/exir/backend/canonical_partitioners/duplicate_dequant_node_pass.py +111 -0
  2360. executorch/exir/backend/canonical_partitioners/group_partitioner.py +371 -0
  2361. executorch/exir/backend/canonical_partitioners/pattern_op_partitioner.py +185 -0
  2362. executorch/exir/backend/compile_spec_schema.py +17 -0
  2363. executorch/exir/backend/operator_support.py +127 -0
  2364. executorch/exir/backend/partitioner.py +114 -0
  2365. executorch/exir/backend/test/backend_with_compiler_demo.py +142 -0
  2366. executorch/exir/backend/test/backend_with_delegate_mapping_demo.py +210 -0
  2367. executorch/exir/backend/test/backend_with_named_data_map.py +115 -0
  2368. executorch/exir/backend/test/backend_with_preprocess_all_demo.py +297 -0
  2369. executorch/exir/backend/test/demo_backend.py +35 -0
  2370. executorch/exir/backend/test/demos/rpc/executor_backend_partitioner.py +77 -0
  2371. executorch/exir/backend/test/demos/rpc/executor_backend_preprocess.py +39 -0
  2372. executorch/exir/backend/test/demos/rpc/test_rpc.py +189 -0
  2373. executorch/exir/backend/test/demos/test_delegate_aten_mode.py +72 -0
  2374. executorch/exir/backend/test/hta_partitioner_demo.py +288 -0
  2375. executorch/exir/backend/test/op_partitioner_demo.py +266 -0
  2376. executorch/exir/backend/test/test_backend_with_named_data_map.py +85 -0
  2377. executorch/exir/backend/test/test_backends.py +1463 -0
  2378. executorch/exir/backend/test/test_backends_lifted.py +1326 -0
  2379. executorch/exir/backend/test/test_backends_nested.py +258 -0
  2380. executorch/exir/backend/test/test_compatibility.py +118 -0
  2381. executorch/exir/backend/test/test_debug_handle_map.py +143 -0
  2382. executorch/exir/backend/test/test_delegate_map_builder.py +257 -0
  2383. executorch/exir/backend/test/test_graph_partition.py +248 -0
  2384. executorch/exir/backend/test/test_group_partitioner.py +1674 -0
  2385. executorch/exir/backend/test/test_lowered_backend_module.py +207 -0
  2386. executorch/exir/backend/test/test_partitioner.py +731 -0
  2387. executorch/exir/backend/test/test_passes.py +48 -0
  2388. executorch/exir/backend/test/test_to_backend_multi_method.py +595 -0
  2389. executorch/exir/backend/test/test_utils.py +264 -0
  2390. executorch/exir/backend/utils.py +661 -0
  2391. executorch/exir/capture/__init__.py +28 -0
  2392. executorch/exir/capture/_capture.py +381 -0
  2393. executorch/exir/capture/_config.py +119 -0
  2394. executorch/exir/capture/_unlift.py +159 -0
  2395. executorch/exir/common.py +140 -0
  2396. executorch/exir/control_flow.py +204 -0
  2397. executorch/exir/debug_handle_utils.py +29 -0
  2398. executorch/exir/delegate.py +171 -0
  2399. executorch/exir/delegate.pyi +21 -0
  2400. executorch/exir/dialects/__init__.py +0 -0
  2401. executorch/exir/dialects/_ops.py +166 -0
  2402. executorch/exir/dialects/backend/_ops.py +123 -0
  2403. executorch/exir/dialects/backend/test/test_backend_ops.py +88 -0
  2404. executorch/exir/dialects/edge/_ops.py +405 -0
  2405. executorch/exir/dialects/edge/arg/model.py +238 -0
  2406. executorch/exir/dialects/edge/arg/type.py +46 -0
  2407. executorch/exir/dialects/edge/dtype/runner.py +122 -0
  2408. executorch/exir/dialects/edge/dtype/supported.py +47 -0
  2409. executorch/exir/dialects/edge/dtype/utils.py +36 -0
  2410. executorch/exir/dialects/edge/edge.yaml +8388 -0
  2411. executorch/exir/dialects/edge/op/api.py +86 -0
  2412. executorch/exir/dialects/edge/op/sample_input.py +1283 -0
  2413. executorch/exir/dialects/edge/op/test/test_api.py +51 -0
  2414. executorch/exir/dialects/edge/spec/gen.py +523 -0
  2415. executorch/exir/dialects/edge/spec/utils.py +328 -0
  2416. executorch/exir/dialects/edge/test/test_edge_ops.py +443 -0
  2417. executorch/exir/dialects/edge/test/test_edge_yaml.py +225 -0
  2418. executorch/exir/dialects/test/test_exir_dialect_ops.py +22 -0
  2419. executorch/exir/dim_order_utils.py +85 -0
  2420. executorch/exir/dynamic_shape.py +12 -0
  2421. executorch/exir/emit/__init__.py +11 -0
  2422. executorch/exir/emit/_emit_program.py +216 -0
  2423. executorch/exir/emit/_emitter.py +2071 -0
  2424. executorch/exir/emit/test/test_emit.py +2405 -0
  2425. executorch/exir/error.py +55 -0
  2426. executorch/exir/graph.py +86 -0
  2427. executorch/exir/graph_module.py +153 -0
  2428. executorch/exir/lowered_backend_module.py +966 -0
  2429. executorch/exir/memory.py +50 -0
  2430. executorch/exir/memory_planning.py +1250 -0
  2431. executorch/exir/operator/convert.py +307 -0
  2432. executorch/exir/operator/manip.py +89 -0
  2433. executorch/exir/operator/test/test_operator.py +84 -0
  2434. executorch/exir/operator/util.py +68 -0
  2435. executorch/exir/pass_base.py +808 -0
  2436. executorch/exir/pass_manager.py +78 -0
  2437. executorch/exir/passes/__init__.py +528 -0
  2438. executorch/exir/passes/_quant_patterns_and_replacements.py +1563 -0
  2439. executorch/exir/passes/const_prop_pass.py +84 -0
  2440. executorch/exir/passes/constant_prop_pass.py +402 -0
  2441. executorch/exir/passes/debug_handle_generator_pass.py +81 -0
  2442. executorch/exir/passes/dim_order_ops_registry.py +90 -0
  2443. executorch/exir/passes/dynamic_shape_prop_pass.py +301 -0
  2444. executorch/exir/passes/executorch_prim_ops_registry.py +166 -0
  2445. executorch/exir/passes/external_constants_pass.py +113 -0
  2446. executorch/exir/passes/init_mutable_pass.py +32 -0
  2447. executorch/exir/passes/insert_write_back_for_buffers_pass.py +185 -0
  2448. executorch/exir/passes/memory_format_ops_pass.py +119 -0
  2449. executorch/exir/passes/memory_planning_pass.py +315 -0
  2450. executorch/exir/passes/normalize_transpose_pass.py +24 -0
  2451. executorch/exir/passes/normalize_view_copy_base_pass.py +62 -0
  2452. executorch/exir/passes/pass_registry.py +89 -0
  2453. executorch/exir/passes/prune_empty_tensors_pass.py +67 -0
  2454. executorch/exir/passes/quant_fusion_pass.py +174 -0
  2455. executorch/exir/passes/quantize_io_pass.py +414 -0
  2456. executorch/exir/passes/reinplace.py +103 -0
  2457. executorch/exir/passes/remove_graph_asserts_pass.py +62 -0
  2458. executorch/exir/passes/remove_mixed_type_operators.py +95 -0
  2459. executorch/exir/passes/remove_noop_pass.py +123 -0
  2460. executorch/exir/passes/remove_unused_parameters_pass.py +55 -0
  2461. executorch/exir/passes/replace_aten_with_edge_pass.py +59 -0
  2462. executorch/exir/passes/replace_broken_ops_with_function_ops_pass.py +42 -0
  2463. executorch/exir/passes/replace_edge_with_backend_pass.py +56 -0
  2464. executorch/exir/passes/replace_sym_size_op_pass.py +35 -0
  2465. executorch/exir/passes/replace_view_copy_with_view_pass.py +325 -0
  2466. executorch/exir/passes/scalar_to_tensor_pass.py +32 -0
  2467. executorch/exir/passes/spec_prop_pass.py +123 -0
  2468. executorch/exir/passes/sym_shape_eval_pass.py +289 -0
  2469. executorch/exir/passes/sym_to_tensor_pass.py +64 -0
  2470. executorch/exir/passes/weights_to_outputs_pass.py +88 -0
  2471. executorch/exir/print_program.py +420 -0
  2472. executorch/exir/program/__init__.py +32 -0
  2473. executorch/exir/program/_fake_program.py +67 -0
  2474. executorch/exir/program/_program.py +1995 -0
  2475. executorch/exir/program/test/__init__.py +0 -0
  2476. executorch/exir/program/test/test_fake_program.py +78 -0
  2477. executorch/exir/program/test/test_program.py +922 -0
  2478. executorch/exir/scalar_type.py +38 -0
  2479. executorch/exir/schema.py +308 -0
  2480. executorch/exir/serde/__init__.py +0 -0
  2481. executorch/exir/serde/export_serialize.py +2872 -0
  2482. executorch/exir/serde/schema.py +406 -0
  2483. executorch/exir/serde/schema_check.py +293 -0
  2484. executorch/exir/serde/serialize.py +838 -0
  2485. executorch/exir/serde/union.py +77 -0
  2486. executorch/exir/sym_util.py +95 -0
  2487. executorch/exir/tensor.py +392 -0
  2488. executorch/exir/tensor_layout.py +39 -0
  2489. executorch/exir/tests/asr_joiner.py +82 -0
  2490. executorch/exir/tests/common.py +96 -0
  2491. executorch/exir/tests/control_flow_models.py +105 -0
  2492. executorch/exir/tests/dynamic_shape_models.py +51 -0
  2493. executorch/exir/tests/models.py +576 -0
  2494. executorch/exir/tests/test_arg_validator.py +72 -0
  2495. executorch/exir/tests/test_capture.py +29 -0
  2496. executorch/exir/tests/test_common.py +81 -0
  2497. executorch/exir/tests/test_delegate.py +308 -0
  2498. executorch/exir/tests/test_dim_order_utils.py +49 -0
  2499. executorch/exir/tests/test_dynamic_shape_propagation.py +174 -0
  2500. executorch/exir/tests/test_error.py +19 -0
  2501. executorch/exir/tests/test_extract_io_quant_params.py +93 -0
  2502. executorch/exir/tests/test_joint_graph.py +124 -0
  2503. executorch/exir/tests/test_memory_format_ops_pass.py +443 -0
  2504. executorch/exir/tests/test_memory_format_ops_pass_aten.py +142 -0
  2505. executorch/exir/tests/test_memory_format_ops_pass_utils.py +266 -0
  2506. executorch/exir/tests/test_memory_planning.py +1135 -0
  2507. executorch/exir/tests/test_op_convert.py +88 -0
  2508. executorch/exir/tests/test_pass_infra.py +179 -0
  2509. executorch/exir/tests/test_passes.py +2332 -0
  2510. executorch/exir/tests/test_print_program.py +54 -0
  2511. executorch/exir/tests/test_prune_empty_tensors_pass.py +77 -0
  2512. executorch/exir/tests/test_quant_fusion_pass.py +528 -0
  2513. executorch/exir/tests/test_quantization.py +120 -0
  2514. executorch/exir/tests/test_quantize_io_pass.py +156 -0
  2515. executorch/exir/tests/test_reinplace_pass.py +124 -0
  2516. executorch/exir/tests/test_remove_unused_parameters_pass.py +205 -0
  2517. executorch/exir/tests/test_remove_view_copy.py +234 -0
  2518. executorch/exir/tests/test_serde.py +352 -0
  2519. executorch/exir/tests/test_tensor.py +341 -0
  2520. executorch/exir/tests/test_tracer.py +497 -0
  2521. executorch/exir/tests/test_verification.py +275 -0
  2522. executorch/exir/tests/test_warnings.py +169 -0
  2523. executorch/exir/tests/transformer.py +122 -0
  2524. executorch/exir/tracer.py +803 -0
  2525. executorch/exir/types.py +18 -0
  2526. executorch/exir/verification/arg_validator.py +140 -0
  2527. executorch/exir/verification/dev_html.py +68 -0
  2528. executorch/exir/verification/interpreter.py +425 -0
  2529. executorch/exir/verification/test/test_verifier.py +177 -0
  2530. executorch/exir/verification/verifier.py +352 -0
  2531. executorch/exir/version.py +9 -0
  2532. executorch/exir/wrap.py +32 -0
  2533. executorch/export/__init__.py +40 -0
  2534. executorch/export/export.py +696 -0
  2535. executorch/export/recipe.py +328 -0
  2536. executorch/export/recipe_provider.py +60 -0
  2537. executorch/export/recipe_registry.py +86 -0
  2538. executorch/export/stages.py +686 -0
  2539. executorch/export/target_recipes.py +185 -0
  2540. executorch/export/tests/test_executorch_export.py +34 -0
  2541. executorch/export/tests/test_export_recipe.py +131 -0
  2542. executorch/export/tests/test_export_session.py +973 -0
  2543. executorch/export/tests/test_export_stages.py +534 -0
  2544. executorch/export/tests/test_recipe_provider.py +98 -0
  2545. executorch/export/tests/test_recipe_registry.py +139 -0
  2546. executorch/export/tests/test_target_recipes.py +539 -0
  2547. executorch/export/types.py +22 -0
  2548. executorch/export/utils.py +51 -0
  2549. executorch/extension/audio/mel_spectrogram.py +279 -0
  2550. executorch/extension/export_util/__init__.py +13 -0
  2551. executorch/extension/export_util/utils.py +147 -0
  2552. executorch/extension/flat_tensor/__init__.py +0 -0
  2553. executorch/extension/flat_tensor/serialize/__init__.py +0 -0
  2554. executorch/extension/flat_tensor/serialize/flat_tensor.fbs +73 -0
  2555. executorch/extension/flat_tensor/serialize/flat_tensor_schema.py +34 -0
  2556. executorch/extension/flat_tensor/serialize/scalar_type.fbs +38 -0
  2557. executorch/extension/flat_tensor/serialize/serialize.py +446 -0
  2558. executorch/extension/flat_tensor/test/test_serialize.py +326 -0
  2559. executorch/extension/gguf_util/convert_main.py +53 -0
  2560. executorch/extension/gguf_util/converter.py +27 -0
  2561. executorch/extension/gguf_util/converters/llama_converter.py +123 -0
  2562. executorch/extension/gguf_util/load_gguf.py +103 -0
  2563. executorch/extension/llm/custom_ops/__init__.py +0 -0
  2564. executorch/extension/llm/custom_ops/custom_ops.py +495 -0
  2565. executorch/extension/llm/custom_ops/custom_ops_aot_lib.dll +0 -0
  2566. executorch/extension/llm/custom_ops/model_sharding.py +117 -0
  2567. executorch/extension/llm/custom_ops/op_tile_crop_aot.py +38 -0
  2568. executorch/extension/llm/custom_ops/preprocess_custom_ops.py +69 -0
  2569. executorch/extension/llm/custom_ops/spinquant/special_hadamard_code_gen.py +279 -0
  2570. executorch/extension/llm/custom_ops/spinquant/third-party/FFHT/example.py +20 -0
  2571. executorch/extension/llm/custom_ops/spinquant/third-party/FFHT/gen.py +869 -0
  2572. executorch/extension/llm/custom_ops/test_preprocess_custom_ops.py +34 -0
  2573. executorch/extension/llm/custom_ops/test_quantized_sdpa.py +550 -0
  2574. executorch/extension/llm/custom_ops/test_sdpa_with_kv_cache.py +651 -0
  2575. executorch/extension/llm/custom_ops/test_update_cache.py +433 -0
  2576. executorch/extension/llm/custom_ops/test_update_cross_attn_cache.py +284 -0
  2577. executorch/extension/llm/export/__init__.py +11 -0
  2578. executorch/extension/llm/export/builder.py +535 -0
  2579. executorch/extension/llm/export/config/llm_config.py +727 -0
  2580. executorch/extension/llm/export/export_llm.py +106 -0
  2581. executorch/extension/llm/export/export_passes.py +202 -0
  2582. executorch/extension/llm/export/partitioner_lib.py +238 -0
  2583. executorch/extension/llm/export/quantizer_lib.py +322 -0
  2584. executorch/extension/llm/export/test/__init__.py +0 -0
  2585. executorch/extension/llm/export/test/test_builder.py +118 -0
  2586. executorch/extension/llm/export/test/test_export_llm.py +155 -0
  2587. executorch/extension/llm/export/test_export_passes.py +209 -0
  2588. executorch/extension/llm/modules/__init__.py +24 -0
  2589. executorch/extension/llm/modules/_position_embeddings.py +729 -0
  2590. executorch/extension/llm/modules/attention.py +415 -0
  2591. executorch/extension/llm/modules/kv_cache.py +151 -0
  2592. executorch/extension/llm/modules/test/__init__.py +0 -0
  2593. executorch/extension/llm/modules/test/test_attention.py +353 -0
  2594. executorch/extension/llm/modules/test/test_kv_cache.py +218 -0
  2595. executorch/extension/llm/modules/test/test_position_embeddings.py +230 -0
  2596. executorch/extension/llm/runner/__init__.py +235 -0
  2597. executorch/extension/llm/runner/_llm_runner.pyi +523 -0
  2598. executorch/extension/llm/runner/test/test_runner_pybindings.py +258 -0
  2599. executorch/extension/llm/tokenizers/pytorch_tokenizers/__init__.py +63 -0
  2600. executorch/extension/llm/tokenizers/pytorch_tokenizers/constants.py +31 -0
  2601. executorch/extension/llm/tokenizers/pytorch_tokenizers/hf_tokenizer.py +57 -0
  2602. executorch/extension/llm/tokenizers/pytorch_tokenizers/llama2c.py +111 -0
  2603. executorch/extension/llm/tokenizers/pytorch_tokenizers/tiktoken.py +198 -0
  2604. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/__init__.py +0 -0
  2605. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/llama2c/__init__.py +0 -0
  2606. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/llama2c/convert.py +49 -0
  2607. executorch/extension/llm/tokenizers/setup.py +199 -0
  2608. executorch/extension/llm/tokenizers/test/test_hf_tokenizer.py +83 -0
  2609. executorch/extension/llm/tokenizers/test/test_python_bindings.py +131 -0
  2610. executorch/extension/llm/tokenizers/test/test_tekken_python.py +453 -0
  2611. executorch/extension/llm/tokenizers/test/test_tiktoken.py +49 -0
  2612. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/abseil.podspec.gen.py +243 -0
  2613. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/copts/copts.py +194 -0
  2614. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/copts/generate_copts.py +109 -0
  2615. executorch/extension/llm/tokenizers/third-party/abseil-cpp/conanfile.py +51 -0
  2616. executorch/extension/llm/tokenizers/third-party/abseil-cpp/create_lts.py +139 -0
  2617. executorch/extension/llm/tokenizers/third-party/json/docs/mkdocs/scripts/check_structure.py +179 -0
  2618. executorch/extension/llm/tokenizers/third-party/json/tests/thirdparty/imapdl/filterbr.py +109 -0
  2619. executorch/extension/llm/tokenizers/third-party/json/tools/amalgamate/amalgamate.py +299 -0
  2620. executorch/extension/llm/tokenizers/third-party/json/tools/gdb_pretty_printer/nlohmann-json.py +32 -0
  2621. executorch/extension/llm/tokenizers/third-party/json/tools/generate_natvis/generate_natvis.py +41 -0
  2622. executorch/extension/llm/tokenizers/third-party/json/tools/serve_header/serve_header.py +410 -0
  2623. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateCommon.py +354 -0
  2624. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateTest.py +186 -0
  2625. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcd.py +999 -0
  2626. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcpHeader.py +98 -0
  2627. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcpTables.py +203 -0
  2628. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateCommon.py +31 -0
  2629. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateDates.py +59 -0
  2630. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateRelease.py +29 -0
  2631. executorch/extension/llm/tokenizers/third-party/re2/benchlog/benchplot.py +98 -0
  2632. executorch/extension/llm/tokenizers/third-party/re2/python/re2.py +583 -0
  2633. executorch/extension/llm/tokenizers/third-party/re2/python/re2_test.py +489 -0
  2634. executorch/extension/llm/tokenizers/third-party/re2/python/setup.py +158 -0
  2635. executorch/extension/llm/tokenizers/third-party/re2/python/toolchains/generate.py +98 -0
  2636. executorch/extension/llm/tokenizers/third-party/re2/re2/make_unicode_casefold.py +151 -0
  2637. executorch/extension/llm/tokenizers/third-party/re2/re2/make_unicode_groups.py +117 -0
  2638. executorch/extension/llm/tokenizers/third-party/re2/re2/unicode.py +284 -0
  2639. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/setup.py +215 -0
  2640. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/__init__.py +1224 -0
  2641. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/_version.py +1 -0
  2642. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/sentencepiece_model_pb2.py +44 -0
  2643. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/sentencepiece_pb2.py +30 -0
  2644. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/test/__init__.py +0 -0
  2645. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/test/sentencepiece_test.py +937 -0
  2646. executorch/extension/module/test/resources/gen_bundled_program.py +99 -0
  2647. executorch/extension/pybindings/_portable_lib.cp313-win_amd64.pyd +0 -0
  2648. executorch/extension/pybindings/portable_lib.py +82 -0
  2649. executorch/extension/pybindings/pybindings.pyi +299 -0
  2650. executorch/extension/pybindings/test/make_test.py +186 -0
  2651. executorch/extension/pybindings/test/test_backend_pybinding.py +27 -0
  2652. executorch/extension/pybindings/test/test_pybindings.py +735 -0
  2653. executorch/extension/pytree/__init__.py +54 -0
  2654. executorch/extension/pytree/test/test.py +331 -0
  2655. executorch/extension/training/__init__.py +22 -0
  2656. executorch/extension/training/examples/CIFAR/data_utils.py +389 -0
  2657. executorch/extension/training/examples/CIFAR/export.py +220 -0
  2658. executorch/extension/training/examples/CIFAR/main.py +188 -0
  2659. executorch/extension/training/examples/CIFAR/model.py +73 -0
  2660. executorch/extension/training/examples/CIFAR/train_utils.py +622 -0
  2661. executorch/extension/training/examples/XOR/export_model.py +74 -0
  2662. executorch/extension/training/examples/XOR/model.py +34 -0
  2663. executorch/extension/training/examples/XOR/test/test_export.py +19 -0
  2664. executorch/extension/training/pybindings/_training_lib.pyi +45 -0
  2665. executorch/extension/training/pybindings/_training_module.py +86 -0
  2666. executorch/extension/training/pybindings/test/test.py +67 -0
  2667. executorch/extension/wasm/test/test_model.py +34 -0
  2668. executorch/include/executorch/extension/kernel_util/make_boxed_from_unboxed_functor.h +260 -0
  2669. executorch/include/executorch/extension/kernel_util/meta_programming.h +119 -0
  2670. executorch/include/executorch/extension/kernel_util/type_list.h +145 -0
  2671. executorch/include/executorch/extension/tensor/tensor.h +14 -0
  2672. executorch/include/executorch/extension/tensor/tensor_accessor.h +218 -0
  2673. executorch/include/executorch/extension/tensor/tensor_ptr.h +485 -0
  2674. executorch/include/executorch/extension/tensor/tensor_ptr_maker.h +687 -0
  2675. executorch/include/executorch/extension/threadpool/cpuinfo_utils.h +24 -0
  2676. executorch/include/executorch/extension/threadpool/threadpool.h +124 -0
  2677. executorch/include/executorch/extension/threadpool/threadpool_guard.h +37 -0
  2678. executorch/include/executorch/runtime/core/array_ref.h +262 -0
  2679. executorch/include/executorch/runtime/core/data_loader.h +141 -0
  2680. executorch/include/executorch/runtime/core/defines.h +20 -0
  2681. executorch/include/executorch/runtime/core/error.h +257 -0
  2682. executorch/include/executorch/runtime/core/evalue.h +608 -0
  2683. executorch/include/executorch/runtime/core/event_tracer.h +592 -0
  2684. executorch/include/executorch/runtime/core/event_tracer_hooks.h +340 -0
  2685. executorch/include/executorch/runtime/core/event_tracer_hooks_delegate.h +204 -0
  2686. executorch/include/executorch/runtime/core/exec_aten/exec_aten.h +175 -0
  2687. executorch/include/executorch/runtime/core/exec_aten/testing_util/tensor_factory.h +1133 -0
  2688. executorch/include/executorch/runtime/core/exec_aten/testing_util/tensor_util.h +378 -0
  2689. executorch/include/executorch/runtime/core/exec_aten/util/dim_order_util.h +278 -0
  2690. executorch/include/executorch/runtime/core/exec_aten/util/scalar_type_util.h +1430 -0
  2691. executorch/include/executorch/runtime/core/exec_aten/util/tensor_dimension_limit.h +21 -0
  2692. executorch/include/executorch/runtime/core/exec_aten/util/tensor_shape_to_c_string.h +68 -0
  2693. executorch/include/executorch/runtime/core/exec_aten/util/tensor_util.h +1415 -0
  2694. executorch/include/executorch/runtime/core/freeable_buffer.h +218 -0
  2695. executorch/include/executorch/runtime/core/function_ref.h +108 -0
  2696. executorch/include/executorch/runtime/core/hierarchical_allocator.h +125 -0
  2697. executorch/include/executorch/runtime/core/memory_allocator.h +233 -0
  2698. executorch/include/executorch/runtime/core/named_data_map.h +77 -0
  2699. executorch/include/executorch/runtime/core/portable_type/bfloat16.h +27 -0
  2700. executorch/include/executorch/runtime/core/portable_type/bfloat16_math.h +14 -0
  2701. executorch/include/executorch/runtime/core/portable_type/bits_types.h +83 -0
  2702. executorch/include/executorch/runtime/core/portable_type/c10/c10/macros/Export.h +1 -0
  2703. executorch/include/executorch/runtime/core/portable_type/c10/c10/macros/Macros.h +1 -0
  2704. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16-inl.h +1 -0
  2705. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16-math.h +299 -0
  2706. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16.h +1 -0
  2707. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/Half-inl.h +1 -0
  2708. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/Half.h +8 -0
  2709. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/TypeSafeSignMath.h +1 -0
  2710. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/bit_cast.h +1 -0
  2711. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex.h +78 -0
  2712. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex_math.h +406 -0
  2713. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex_utils.h +46 -0
  2714. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/floating_point_utils.h +1 -0
  2715. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/irange.h +123 -0
  2716. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/llvmMathExtras.h +905 -0
  2717. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/overflows.h +100 -0
  2718. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/safe_numerics.h +114 -0
  2719. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Export.h +153 -0
  2720. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h +694 -0
  2721. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/BFloat16.h +480 -0
  2722. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/Half.h +788 -0
  2723. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/TypeSafeSignMath.h +148 -0
  2724. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/bit_cast.h +50 -0
  2725. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/complex.h +616 -0
  2726. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/floating_point_utils.h +38 -0
  2727. executorch/include/executorch/runtime/core/portable_type/complex.h +21 -0
  2728. executorch/include/executorch/runtime/core/portable_type/device.h +74 -0
  2729. executorch/include/executorch/runtime/core/portable_type/half.h +27 -0
  2730. executorch/include/executorch/runtime/core/portable_type/optional.h +36 -0
  2731. executorch/include/executorch/runtime/core/portable_type/qint_types.h +83 -0
  2732. executorch/include/executorch/runtime/core/portable_type/scalar.h +123 -0
  2733. executorch/include/executorch/runtime/core/portable_type/scalar_type.h +154 -0
  2734. executorch/include/executorch/runtime/core/portable_type/string_view.h +29 -0
  2735. executorch/include/executorch/runtime/core/portable_type/tensor.h +175 -0
  2736. executorch/include/executorch/runtime/core/portable_type/tensor_impl.h +304 -0
  2737. executorch/include/executorch/runtime/core/portable_type/tensor_options.h +60 -0
  2738. executorch/include/executorch/runtime/core/result.h +268 -0
  2739. executorch/include/executorch/runtime/core/span.h +110 -0
  2740. executorch/include/executorch/runtime/core/tag.h +90 -0
  2741. executorch/include/executorch/runtime/core/tensor_layout.h +93 -0
  2742. executorch/include/executorch/runtime/core/tensor_shape_dynamism.h +39 -0
  2743. executorch/include/executorch/runtime/core/testing_util/error_matchers.h +303 -0
  2744. executorch/include/executorch/runtime/kernel/kernel_includes.h +23 -0
  2745. executorch/include/executorch/runtime/kernel/kernel_runtime_context.h +131 -0
  2746. executorch/include/executorch/runtime/kernel/operator_registry.h +298 -0
  2747. executorch/include/executorch/runtime/kernel/test/test_util.h +40 -0
  2748. executorch/include/executorch/runtime/kernel/thread_parallel_interface.h +107 -0
  2749. executorch/include/executorch/runtime/platform/abort.h +36 -0
  2750. executorch/include/executorch/runtime/platform/assert.h +124 -0
  2751. executorch/include/executorch/runtime/platform/clock.h +43 -0
  2752. executorch/include/executorch/runtime/platform/compat_unistd.h +75 -0
  2753. executorch/include/executorch/runtime/platform/compiler.h +214 -0
  2754. executorch/include/executorch/runtime/platform/log.h +216 -0
  2755. executorch/include/executorch/runtime/platform/platform.h +276 -0
  2756. executorch/include/executorch/runtime/platform/profiler.h +289 -0
  2757. executorch/include/executorch/runtime/platform/runtime.h +35 -0
  2758. executorch/include/executorch/runtime/platform/system.h +49 -0
  2759. executorch/include/executorch/runtime/platform/test/pal_spy.h +86 -0
  2760. executorch/include/executorch/runtime/platform/test/stub_platform.h +76 -0
  2761. executorch/include/executorch/runtime/platform/types.h +24 -0
  2762. executorch/kernels/quantized/__init__.py +21 -0
  2763. executorch/kernels/quantized/quantized.yaml +113 -0
  2764. executorch/kernels/quantized/quantized_ops_aot_lib.dll +0 -0
  2765. executorch/kernels/quantized/test/supported_features_def.yaml +1 -0
  2766. executorch/kernels/quantized/test/test_out_variants.py +100 -0
  2767. executorch/kernels/quantized/test/test_quant_dequant_per_token.py +149 -0
  2768. executorch/runtime/__init__.py +334 -0
  2769. executorch/runtime/kernel/test/functions.yaml +13 -0
  2770. executorch/runtime/test/test_runtime.py +106 -0
  2771. executorch/runtime/test/test_runtime_etdump_gen.py +163 -0
  2772. executorch/schema/program.fbs +489 -0
  2773. executorch/schema/scalar_type.fbs +43 -0
  2774. executorch/share/cmake/executorch-config.cmake +77 -0
  2775. executorch/util/activation_memory_profiler.py +177 -0
  2776. executorch/util/collect_env.py +748 -0
  2777. executorch/util/python_profiler.py +96 -0
  2778. executorch/version.py +4 -0
  2779. executorch-1.1.0.dist-info/METADATA +87 -0
  2780. executorch-1.1.0.dist-info/RECORD +2784 -0
  2781. executorch-1.1.0.dist-info/WHEEL +5 -0
  2782. executorch-1.1.0.dist-info/entry_points.txt +2 -0
  2783. executorch-1.1.0.dist-info/licenses/LICENSE +37 -0
  2784. executorch-1.1.0.dist-info/top_level.txt +7 -0
@@ -0,0 +1,2872 @@
1
+ # Copyright (c) Meta Platforms, Inc. and affiliates.
2
+ # All rights reserved.
3
+ #
4
+ # This source code is licensed under the BSD-style license found in the
5
+ # LICENSE file in the root directory of this source tree.
6
+
7
+ # pyre-ignore-all-errors
8
+
9
+ # Copied over from caffe2/torch/_export/serde/serialize.py until dialects
10
+ # are supported in torch export serializer.
11
+
12
+ import base64
13
+ import copy
14
+ import copyreg
15
+ import dataclasses
16
+ import heapq
17
+ import inspect
18
+ import io
19
+ import json
20
+ import logging
21
+ import math
22
+ import operator
23
+ import re
24
+ import typing
25
+
26
+ from contextlib import contextmanager
27
+ from dataclasses import dataclass, field
28
+ from enum import Enum
29
+ from typing import (
30
+ Any,
31
+ Callable,
32
+ cast,
33
+ Dict,
34
+ final,
35
+ Iterator,
36
+ List,
37
+ Optional,
38
+ Set,
39
+ Tuple,
40
+ Union,
41
+ )
42
+
43
+ import sympy
44
+
45
+ import torch
46
+ import torch.export.exported_program
47
+ import torch.export.exported_program as ep
48
+ from torch._export.serde.schema import SchemaVersion
49
+ from torch._export.verifier import load_verifier
50
+ from torch._subclasses.fake_tensor import FakeTensor, FakeTensorMode
51
+ from torch.fx.experimental import symbolic_shapes
52
+ from torch.utils import _pytree as pytree
53
+ from torch.utils._pytree import treespec_dumps, treespec_loads
54
+ from torch.utils._sympy.numbers import int_oo
55
+ from torch.utils._sympy.value_ranges import ValueRanges
56
+
57
+ import executorch.exir as exir
58
+
59
+ # pyre-ignore
60
+
61
+ from .schema import ( # type: ignore[attr-defined]
62
+ Argument,
63
+ BufferMutationSpec,
64
+ ConstantInputSpec,
65
+ ConstantValue,
66
+ CustomObjArgument,
67
+ Device,
68
+ ExportedProgram,
69
+ GradientToParameterSpec,
70
+ GradientToUserInputSpec,
71
+ Graph,
72
+ GraphArgument,
73
+ GraphModule,
74
+ GraphSignature,
75
+ InputSpec,
76
+ InputToBufferSpec,
77
+ InputToCustomObjSpec,
78
+ InputTokenSpec,
79
+ InputToParameterSpec,
80
+ InputToTensorConstantSpec,
81
+ Layout,
82
+ LossOutputSpec,
83
+ MemoryFormat,
84
+ ModuleCallEntry,
85
+ ModuleCallSignature,
86
+ NamedArgument,
87
+ Node,
88
+ OptionalTensorArgument,
89
+ OutputSpec,
90
+ OutputTokenSpec,
91
+ RangeConstraint,
92
+ ScalarType,
93
+ SCHEMA_VERSION,
94
+ SymBool,
95
+ SymBoolArgument,
96
+ SymExpr,
97
+ SymExprHint,
98
+ SymInt,
99
+ SymIntArgument,
100
+ TensorArgument,
101
+ TensorMeta,
102
+ TokenArgument,
103
+ TREESPEC_VERSION,
104
+ UserInputMutationSpec,
105
+ UserInputSpec,
106
+ UserOutputSpec,
107
+ )
108
+ from .union import _Union
109
+
110
+
111
+ __all__ = [
112
+ "serialize",
113
+ "GraphModuleSerializer",
114
+ "ExportedProgramSerializer",
115
+ "GraphModuleDeserializer",
116
+ "ExportedProgramDeserializer",
117
+ ]
118
+
119
+
120
+ log = logging.getLogger(__name__)
121
+
122
+
123
+ class SerializeError(RuntimeError):
124
+ pass
125
+
126
+
127
+ def _reverse_map(d: Dict[Any, Enum]):
128
+ return {v.value: k for k, v in d.items()}
129
+
130
+
131
+ MetaType = Union[
132
+ FakeTensor, int, torch.SymInt, bool, torch.SymBool, ep.CustomObjArgument
133
+ ]
134
+
135
+
136
+ ST_DELIMITER = ";"
137
+
138
+ _TORCH_TO_SERIALIZE_DTYPE = {
139
+ torch.uint8: ScalarType.BYTE,
140
+ torch.int8: ScalarType.CHAR,
141
+ torch.int16: ScalarType.SHORT,
142
+ torch.int32: ScalarType.INT,
143
+ torch.int64: ScalarType.LONG,
144
+ torch.float16: ScalarType.HALF,
145
+ torch.float32: ScalarType.FLOAT,
146
+ torch.float64: ScalarType.DOUBLE,
147
+ torch.complex32: ScalarType.COMPLEXHALF,
148
+ torch.complex64: ScalarType.COMPLEXFLOAT,
149
+ torch.complex128: ScalarType.COMPLEXDOUBLE,
150
+ torch.bool: ScalarType.BOOL,
151
+ torch.bfloat16: ScalarType.BFLOAT16,
152
+ torch.uint16: ScalarType.UINT16
153
+ }
154
+
155
+
156
+ _SERIALIZE_TO_TORCH_DTYPE = _reverse_map(_TORCH_TO_SERIALIZE_DTYPE) # type: ignore[arg-type]
157
+
158
+
159
+ _TORCH_TO_SERIALIZE_LAYOUT = {
160
+ torch.sparse_coo: Layout.SparseCoo,
161
+ torch.sparse_csr: Layout.SparseCsr,
162
+ torch.sparse_csc: Layout.SparseCsc,
163
+ torch.sparse_bsr: Layout.SparseBsr,
164
+ torch.sparse_bsc: Layout.SparseBsc,
165
+ torch._mkldnn: Layout._mkldnn, # type: ignore[attr-defined]
166
+ torch.strided: Layout.Strided,
167
+ }
168
+
169
+
170
+ _SERIALIZE_TO_TORCH_LAYOUT = _reverse_map(_TORCH_TO_SERIALIZE_LAYOUT) # type: ignore[arg-type]
171
+
172
+
173
+ _TORCH_TO_SERIALIZE_MEMORY_FORMAT = {
174
+ torch.contiguous_format: MemoryFormat.ContiguousFormat,
175
+ torch.channels_last: MemoryFormat.ChannelsLast,
176
+ torch.channels_last_3d: MemoryFormat.ChannelsLast3d,
177
+ torch.preserve_format: MemoryFormat.PreserveFormat,
178
+ }
179
+
180
+
181
+ _SERIALIZE_TO_TORCH_MEMORY_FORMAT = _reverse_map(_TORCH_TO_SERIALIZE_MEMORY_FORMAT) # type: ignore[arg-type]
182
+
183
+
184
+ _SYM_INT_OPS = {
185
+ operator.mul,
186
+ operator.add,
187
+ operator.sub,
188
+ operator.floordiv,
189
+ operator.mod,
190
+ torch.sym_int,
191
+ torch.sym_float,
192
+ torch.sym_ite,
193
+ torch.sym_max,
194
+ torch.sym_min,
195
+ torch.sym_sqrt,
196
+ }
197
+
198
+
199
+ _SYM_BOOL_OPS = {
200
+ operator.eq,
201
+ operator.ne,
202
+ operator.le,
203
+ operator.ge,
204
+ operator.lt,
205
+ operator.gt,
206
+ torch.sym_not,
207
+ }
208
+
209
+
210
+ _KNOWN_FUNCTIONS = {
211
+ exir.memory.view,
212
+ }
213
+
214
+
215
+ @dataclass
216
+ class SerializedArtifact:
217
+ exported_program: bytes
218
+ state_dict: bytes
219
+ constants: bytes
220
+ example_inputs: bytes
221
+
222
+
223
+ @dataclass
224
+ class _SerializedProgram:
225
+ exported_program: ExportedProgram
226
+ state_dict: bytes
227
+ constants: bytes
228
+ example_inputs: bytes
229
+
230
+
231
+ def deserialize_device(d: Device) -> torch.device:
232
+ if d.index is None:
233
+ return torch.device(type=d.type) # type: ignore[call-overload]
234
+ return torch.device(type=d.type, index=d.index)
235
+
236
+
237
+ def serialize_sym_int(s: Union[int, torch.SymInt]) -> SymInt:
238
+ if isinstance(s, (torch.SymInt, int)):
239
+ if symbolic_shapes.is_concrete_int(s):
240
+ return SymInt.create(as_int=int(s))
241
+ else:
242
+ assert isinstance(s, torch.SymInt)
243
+ if s.node.hint is None:
244
+ return SymInt.create(as_expr=SymExpr(str(s)))
245
+ else:
246
+ return SymInt.create(
247
+ as_expr=SymExpr(str(s), hint=SymExprHint.create(as_int=s.node.hint))
248
+ )
249
+ else:
250
+ raise SerializeError(
251
+ f"SymInt should be either symbol or int, got `{s}` of type `{type(s)}`"
252
+ )
253
+
254
+
255
+ def serialize_sym_bool(s: Union[bool, torch.SymBool]) -> SymBool:
256
+ if isinstance(s, (torch.SymBool, bool)):
257
+ if symbolic_shapes.is_concrete_bool(s):
258
+ return SymBool.create(as_bool=bool(s))
259
+ else:
260
+ return SymBool.create(as_expr=SymExpr(expr_str=str(s)))
261
+ else:
262
+ raise SerializeError(
263
+ f"SymBool should be either symbol or bool, got `{s}` of type `{type(s)}`"
264
+ )
265
+
266
+
267
+ def serialize_tensor_meta(t: torch.Tensor) -> TensorMeta:
268
+ """
269
+ Extract a TensorMeta describing `t`.
270
+ """
271
+ return TensorMeta(
272
+ dtype=_TORCH_TO_SERIALIZE_DTYPE[t.dtype],
273
+ sizes=[serialize_sym_int(s) for s in t.shape],
274
+ requires_grad=t.requires_grad,
275
+ device=Device(type=t.device.type, index=t.device.index),
276
+ strides=[serialize_sym_int(s) for s in t.stride()],
277
+ storage_offset=serialize_sym_int(0), # TODO needs to be fixed.
278
+ layout=_TORCH_TO_SERIALIZE_LAYOUT[t.layout],
279
+ )
280
+
281
+
282
+ _CURRENT_DESERIALIZER: List["GraphModuleDeserializer"] = []
283
+
284
+
285
+ def _reduce_fake_tensor(fake_tensor: FakeTensor):
286
+ is_parameter = isinstance(fake_tensor, torch.nn.Parameter)
287
+ tensor_meta = serialize_tensor_meta(fake_tensor)
288
+ tensor_meta_bytes = json.dumps(
289
+ _dataclass_to_dict(tensor_meta), cls=EnumEncoder
290
+ ).encode("utf-8")
291
+ return _reconstruct_fake_tensor, (tensor_meta_bytes, is_parameter)
292
+
293
+
294
+ def _reconstruct_fake_tensor(
295
+ serialized_tensor_meta: bytes, is_parameter: bool
296
+ ) -> FakeTensor:
297
+ # Deserialize the bytes into a TensorMeta
298
+ json_tensor_meta = json.loads(serialized_tensor_meta.decode("utf-8"))
299
+ tensor_meta = _dict_to_dataclass(TensorMeta, json_tensor_meta)
300
+ # Find the current fake mode
301
+ assert len(_CURRENT_DESERIALIZER) != 0, "Need access to current deserializer state"
302
+ fake_tensor = _CURRENT_DESERIALIZER[-1].deserialize_tensor_meta(tensor_meta)
303
+ if is_parameter:
304
+ fake_tensor = torch.nn.Parameter(fake_tensor) # type: ignore[assignment]
305
+ return fake_tensor
306
+
307
+
308
+ def serialize_torch_artifact(artifact: Dict[str, Any]) -> bytes:
309
+ assert (
310
+ FakeTensor not in copyreg.dispatch_table
311
+ ), "Refusing to stomp on existing FakeTensor reducer"
312
+ try:
313
+ copyreg.pickle(FakeTensor, _reduce_fake_tensor)
314
+ buffer = io.BytesIO()
315
+ # This is a workaround for backend's tensor deserialization problem:
316
+ # unpickleTensor() always create a tensor on the device where it was originally saved
317
+ # This behavior is bad for multi-gpu training, as we wish to directly load the tensor
318
+ # on the designated device.
319
+ # For now, we simply move the tensor to cpu before saving.
320
+ # TODO: this should be fixed by deserialization instead.
321
+ torch.save(artifact, buffer)
322
+ return buffer.getvalue()
323
+ finally:
324
+ del copyreg.dispatch_table[FakeTensor]
325
+
326
+
327
+ def deserialize_torch_artifact(
328
+ serialized: Union[Dict[str, Any], Tuple[Any, ...], bytes]
329
+ ):
330
+ if isinstance(serialized, (dict, tuple)):
331
+ return serialized
332
+ if len(serialized) == 0:
333
+ return {}
334
+ buffer = io.BytesIO(serialized)
335
+ buffer.seek(0)
336
+ artifact = torch.load(buffer)
337
+ assert isinstance(artifact, (tuple, dict))
338
+ return artifact
339
+
340
+
341
+ def _sympy_int_to_int(val: sympy.Expr, adjust: str):
342
+ # Convert simple sympy Integers into concrete int
343
+ if val in (sympy.oo, int_oo):
344
+ return math.inf
345
+ if val in (-sympy.oo, -int_oo):
346
+ return -math.inf
347
+ if isinstance(val, sympy.Integer):
348
+ return int(val)
349
+
350
+ # TODO: Remove this adjustment when Ed gets rid of fractional ranges
351
+ log.warning(
352
+ "Export constraints cannot be non-integer expressions. Found "
353
+ "type %s, and value %s. We will attempt to %s "
354
+ "this value.",
355
+ type(val),
356
+ val,
357
+ adjust,
358
+ )
359
+
360
+ if adjust == "floor":
361
+ return math.floor(val)
362
+ elif adjust == "ceil":
363
+ return math.ceil(val)
364
+ else:
365
+ raise RuntimeError(f"Got invalid adjustment {adjust}")
366
+
367
+
368
+ def _int_to_sympy_int(val) -> sympy.Expr:
369
+ # Convert concrete int into simple sympy Integers
370
+ if val == math.inf:
371
+ return int_oo
372
+ if val == -math.inf:
373
+ return -int_oo
374
+ return sympy.Integer(val)
375
+
376
+
377
+ def serialize_range_constraints(
378
+ range_constraints: Dict[sympy.Symbol, ValueRanges]
379
+ ) -> Dict[str, RangeConstraint]:
380
+ return {
381
+ str(k): RangeConstraint(
382
+ _sympy_int_to_int(v.lower, "ceil"), # type: ignore[arg-type]
383
+ _sympy_int_to_int(v.upper, "floor"), # type: ignore[arg-type]
384
+ )
385
+ for k, v in range_constraints.items()
386
+ }
387
+
388
+
389
+ def _is_single_tensor_return(target: torch._ops.OpOverload) -> bool:
390
+ returns = target._schema.returns
391
+ return len(returns) == 1 and isinstance(returns[0].real_type, torch.TensorType)
392
+
393
+
394
+ def _is_single_tensor_list_return(target: torch._ops.OpOverload) -> bool:
395
+ returns = target._schema.returns
396
+ if len(returns) != 1:
397
+ return False
398
+ return_type = returns[0].real_type
399
+ return isinstance(return_type, torch.ListType) and isinstance(
400
+ return_type.getElementType(), torch.TensorType
401
+ )
402
+
403
+
404
+ def _output_node_at_index(node, index):
405
+ for user in node.users:
406
+ assert user.target is operator.getitem, f"{user} is not a getitem node"
407
+ if index == user.args[1]:
408
+ return user
409
+ return None
410
+
411
+
412
+ @dataclass
413
+ class GraphState:
414
+ inputs: List[Argument] = field(default_factory=list)
415
+ outputs: List[Argument] = field(default_factory=list)
416
+ nodes: List[Node] = field(default_factory=list)
417
+ tensor_values: Dict[str, TensorMeta] = field(default_factory=dict)
418
+ sym_int_values: Dict[str, SymInt] = field(default_factory=dict)
419
+ sym_bool_values: Dict[str, SymBool] = field(default_factory=dict)
420
+ is_single_tensor_return: bool = False
421
+ custom_obj_values: Dict[str, CustomObjArgument] = field(default_factory=dict)
422
+
423
+
424
+ class Final(type):
425
+ def __new__(metacls, name, bases, classdict):
426
+ for b in bases:
427
+ if isinstance(b, Final):
428
+ raise TypeError(f"type '{b.__name__}' is not an acceptable base type")
429
+ return type.__new__(metacls, name, bases, dict(classdict))
430
+
431
+
432
+ class GraphModuleSerializer:
433
+ def __init__(
434
+ self,
435
+ graph_signature: ep.ExportGraphSignature,
436
+ module_call_graph: List[ep.ModuleCallEntry],
437
+ ):
438
+ self.graph_state = GraphState()
439
+ self.graph_signature = graph_signature
440
+ self.module_call_graph = module_call_graph
441
+ self.custom_objs: Dict[str, torch._C.ScriptObject] = {}
442
+
443
+ @contextmanager
444
+ def save_graph_state(self):
445
+ saved = self.graph_state
446
+ self.graph_state = GraphState()
447
+ try:
448
+ yield
449
+ finally:
450
+ self.graph_state = saved
451
+
452
+ def handle_placeholder(self, node: torch.fx.Node):
453
+ assert node.op == "placeholder"
454
+ if isinstance(node.meta["val"], torch.Tensor):
455
+ graph_input = Argument.create(as_tensor=TensorArgument(name=node.name))
456
+ self.graph_state.tensor_values[node.name] = serialize_tensor_meta(
457
+ node.meta["val"]
458
+ )
459
+ elif isinstance(node.meta["val"], torch.SymInt):
460
+ graph_input = Argument.create(
461
+ as_sym_int=SymIntArgument.create(as_name=node.name)
462
+ )
463
+ self.graph_state.sym_int_values[node.name] = serialize_sym_int(
464
+ node.meta["val"]
465
+ )
466
+ elif isinstance(node.meta["val"], (int, bool, str, float, type(None))):
467
+ graph_input = self.serialize_input(node.meta["val"])
468
+ elif isinstance(node.meta["val"], ep.CustomObjArgument):
469
+ class_fqn = node.meta["val"].class_fqn
470
+ graph_input = Argument.create(
471
+ as_custom_obj=CustomObjArgument(name=node.name, class_fqn=class_fqn)
472
+ )
473
+ self.graph_state.custom_obj_values[node.name] = (
474
+ self.serialize_script_obj_meta(node.meta["val"])
475
+ )
476
+ else:
477
+ raise AssertionError(f"Unimplemented graph input type: {node.meta['val']}")
478
+ self.graph_state.inputs.append(graph_input)
479
+
480
+ def handle_output(self, node: torch.fx.Node):
481
+ assert node.op == "output"
482
+ assert len(node.args) == 1, "FX.Node's args should have one arg"
483
+ node_args = node.args[0]
484
+ if isinstance(node_args, torch.fx.Node):
485
+ # For singleton tensor returns
486
+ self.graph_state.is_single_tensor_return = True
487
+ self.graph_state.outputs = [self.serialize_input(node_args)]
488
+ else:
489
+ assert isinstance(node_args, (tuple, list))
490
+ self.graph_state.outputs = [self.serialize_input(arg) for arg in node_args]
491
+
492
+ def serialize_operator(self, target) -> str:
493
+ if isinstance(target, str):
494
+ return target
495
+ elif target.__module__.startswith("torch._ops"):
496
+ # TODO(zhxchen17) Maybe provide a function name helper in FX.
497
+ # From torch.fx.node._get_qualified_name
498
+ module = target.__module__.replace("torch._ops", "torch.ops")
499
+ return f"{module}.{target.__name__}"
500
+ else: # TODO(zhxchen17) Don't catch all here.
501
+ return f"{target.__module__}.{target.__name__}"
502
+
503
+ def handle_call_function(self, node: torch.fx.Node):
504
+ assert node.op == "call_function"
505
+
506
+ # getitem has been handled in the producer node, skip it here
507
+ if node.target is operator.getitem:
508
+ return
509
+
510
+ if node.target in _SYM_INT_OPS:
511
+ assert len(node.kwargs) == 0
512
+ meta_val = node.meta["val"]
513
+ ex_node = Node(
514
+ name=node.name,
515
+ target=self.serialize_operator(node.target),
516
+ inputs=self.serialize_sym_op_inputs(node.target, node.args),
517
+ outputs=[
518
+ Argument.create(
519
+ as_sym_int=self.serialize_sym_int_output(node.name, meta_val)
520
+ )
521
+ ],
522
+ metadata=self.serialize_metadata(node),
523
+ )
524
+ elif node.target in _SYM_BOOL_OPS:
525
+ assert len(node.kwargs) == 0
526
+ meta_val = node.meta["val"]
527
+ ex_node = Node(
528
+ name=node.name,
529
+ target=self.serialize_operator(node.target),
530
+ inputs=self.serialize_sym_op_inputs(node.target, node.args),
531
+ outputs=[
532
+ Argument.create(
533
+ as_sym_bool=self.serialize_sym_bool_output(node.name, meta_val)
534
+ )
535
+ ],
536
+ metadata=self.serialize_metadata(node),
537
+ )
538
+ elif isinstance(node.target, torch._ops.OpOverload):
539
+ ex_node = Node(
540
+ name=node.name,
541
+ target=self.serialize_operator(node.target),
542
+ inputs=self.serialize_inputs(node.target, node.args, node.kwargs),
543
+ outputs=self.serialize_outputs(node),
544
+ # TODO: create a new tensor_values here, meta might have faketensor info
545
+ metadata=self.serialize_metadata(node),
546
+ )
547
+ elif isinstance(node.target, torch._ops.HigherOrderOperator):
548
+ ex_node = Node(
549
+ name=node.name,
550
+ target=self.serialize_operator(node.target),
551
+ inputs=self.serialize_hoo_inputs(node.args, node.kwargs),
552
+ outputs=self.serialize_hoo_outputs(node),
553
+ metadata=self.serialize_metadata(node),
554
+ )
555
+ elif node.target in _KNOWN_FUNCTIONS:
556
+ ex_node = Node(
557
+ name=node.name,
558
+ target=node._pretty_print_target(node.target),
559
+ inputs=self.serialize_hoo_inputs(node.args, node.kwargs),
560
+ outputs=self.serialize_hoo_outputs(node),
561
+ metadata=self.serialize_metadata(node),
562
+ )
563
+ else:
564
+ raise SerializeError(f"Serializing {node.target} is not supported")
565
+
566
+ self.graph_state.nodes.append(ex_node)
567
+
568
+ def handle_get_attr(self, node):
569
+ pass
570
+
571
+ def serialize_metadata(self, node: torch.fx.Node) -> Dict[str, str]:
572
+ ret = {}
573
+ if stack_trace := node.meta.get("stack_trace"):
574
+ ret["stack_trace"] = stack_trace
575
+
576
+ if nn_module_stack := node.meta.get("nn_module_stack"):
577
+
578
+ def export_nn_module_stack(val):
579
+ assert isinstance(val, tuple) and len(val) == 2
580
+ path, ty = val
581
+
582
+ assert isinstance(path, str)
583
+
584
+ # node.meta["nn_module_stack"] could have two forms:
585
+ # 1. (path: str, module_type: 'type'), e.g.
586
+ # ('', <class 'sigmoid.inference.MySimpleModel'>)
587
+ # 2. (path: str, module_type: str), e.g.
588
+ # ('', 'sigmoid.inference.MySimpleModel')
589
+ # ExportedProgram directly produced by torch.export() has form 1
590
+ # ExportedProgram deserialized from disk has form 2
591
+ # TODO: This is not ideal, we should fix this.
592
+ if isinstance(ty, str):
593
+ normalized_ty = ty
594
+ else:
595
+ normalized_ty = ty.__module__ + "." + ty.__qualname__
596
+
597
+ return path + "," + normalized_ty
598
+
599
+ # Serialize to "key,orig_path,type_str"
600
+ nn_module_list = [
601
+ f"{k},{export_nn_module_stack(v)}" for k, v in nn_module_stack.items()
602
+ ]
603
+ ret["nn_module_stack"] = ST_DELIMITER.join(nn_module_list)
604
+
605
+ if source_fn_st := node.meta.get("source_fn_stack"):
606
+ source_fn_list = [
607
+ f"{source_fn[0]},{self.serialize_operator(source_fn[1])}"
608
+ for source_fn in source_fn_st
609
+ ]
610
+ ret["source_fn_stack"] = ST_DELIMITER.join(source_fn_list)
611
+
612
+ if torch_fn := node.meta.get("torch_fn"):
613
+ ret["torch_fn"] = ST_DELIMITER.join(list(torch_fn))
614
+
615
+ return ret
616
+
617
+ def serialize_script_obj_meta(
618
+ self, script_obj_meta: ep.CustomObjArgument
619
+ ) -> CustomObjArgument:
620
+ return CustomObjArgument(
621
+ name=script_obj_meta.name,
622
+ class_fqn=script_obj_meta.class_fqn,
623
+ )
624
+
625
+ def serialize_sym_op_inputs(self, op, args) -> List[NamedArgument]:
626
+ serialized_args = []
627
+ args_names = inspect.signature(op).parameters.keys()
628
+ for args_name, arg in zip(args_names, args):
629
+ serialized_args.append(
630
+ NamedArgument(name=args_name, arg=self.serialize_input(arg))
631
+ )
632
+ return serialized_args
633
+
634
+ def serialize_inputs(
635
+ self, target: torch._ops.OpOverload, args, kwargs=None
636
+ ) -> List[NamedArgument]:
637
+ assert isinstance(target, torch._ops.OpOverload)
638
+ kwargs = kwargs or {}
639
+ serialized_args = []
640
+ for i, schema_arg in enumerate(target._schema.arguments):
641
+ if schema_arg.name in kwargs:
642
+ serialized_args.append(
643
+ NamedArgument(
644
+ name=schema_arg.name,
645
+ arg=self.serialize_input(
646
+ kwargs[schema_arg.name], schema_arg.type
647
+ ),
648
+ )
649
+ )
650
+ elif not schema_arg.kwarg_only and i < len(args):
651
+ serialized_args.append(
652
+ NamedArgument(
653
+ name=schema_arg.name,
654
+ arg=self.serialize_input(args[i], schema_arg.type),
655
+ )
656
+ )
657
+ else:
658
+ # We intentionally don't serialize the missing arguments
659
+ # with default values
660
+ pass
661
+
662
+ return serialized_args
663
+
664
+ def serialize_hoo_inputs(self, args, kwargs) -> List[NamedArgument]:
665
+ """
666
+ For serializing HOO inputs since HOOs do not have a schema.
667
+ """
668
+ inputs = [
669
+ NamedArgument(
670
+ name="",
671
+ arg=self.serialize_input(a),
672
+ )
673
+ for a in args
674
+ ]
675
+ inputs.extend(
676
+ [
677
+ NamedArgument(name=name, arg=self.serialize_input(a))
678
+ for name, a in kwargs.items()
679
+ ]
680
+ )
681
+ return inputs
682
+
683
+ def is_sym_int_arg(self, arg) -> bool:
684
+ return isinstance(arg, int) or (
685
+ isinstance(arg, torch.fx.Node)
686
+ and arg.name in self.graph_state.sym_int_values
687
+ )
688
+
689
+ def is_sym_bool_arg(self, arg) -> bool:
690
+ return isinstance(arg, bool) or (
691
+ isinstance(arg, torch.fx.Node)
692
+ and arg.name in self.graph_state.sym_bool_values
693
+ )
694
+
695
+ def serialize_input(
696
+ self, arg, arg_type: Optional[torch._C.Argument] = None
697
+ ) -> Argument:
698
+ import torch._inductor.ir as inductor_ir
699
+
700
+ inductor_tensor_buffers = (
701
+ inductor_ir.Buffer,
702
+ inductor_ir.ReinterpretView,
703
+ )
704
+
705
+ if isinstance(arg, torch.fx.Node):
706
+ if arg.op == "get_attr":
707
+ assert isinstance(arg.target, str)
708
+ attr = getattr(arg.graph.owning_module, arg.target)
709
+
710
+ if isinstance(attr, torch.Tensor):
711
+ raise SerializeError(
712
+ "getattr nodes containing tensors should not appear in the graph"
713
+ )
714
+ elif isinstance(attr, torch.fx.GraphModule):
715
+ with self.save_graph_state():
716
+ graph = self.serialize_graph(attr)
717
+ return Argument.create(
718
+ as_graph=GraphArgument(name=arg.target, graph=graph)
719
+ )
720
+ else:
721
+ raise SerializeError(
722
+ f"Unsupported getattr attribute {arg.target} with type: {type(attr)}"
723
+ )
724
+ elif self.is_sym_int_arg(arg):
725
+ return Argument.create(
726
+ as_sym_int=SymIntArgument.create(as_name=arg.name)
727
+ )
728
+ elif self.is_sym_bool_arg(arg):
729
+ return Argument.create(
730
+ as_sym_bool=SymBoolArgument.create(as_name=arg.name)
731
+ )
732
+ else:
733
+ if isinstance(arg.meta["val"], ep.CustomObjArgument):
734
+ return Argument.create(
735
+ as_custom_obj=CustomObjArgument(
736
+ name=arg.name, class_fqn=arg.meta["val"].class_fqn
737
+ )
738
+ )
739
+ return Argument.create(as_tensor=TensorArgument(name=arg.name))
740
+ elif isinstance(arg, inductor_tensor_buffers):
741
+ # Other branches are for arguments in fx node.
742
+ # This is a special branch for handling buffers (representing tensor arguments)
743
+ # for inductor's ExternalFallbackNode
744
+ # export_extern_kernel_node() is using this function to serialize arguments
745
+ arg_name = arg.get_name()
746
+ assert arg_name is not None, "Buffer must have valid name"
747
+ return Argument.create(as_tensor=TensorArgument(name=arg_name))
748
+ elif isinstance(arg, torch.SymInt):
749
+ # This is a special branch for handling SymInt args in inductor's
750
+ # ExternalFallbackNode.
751
+ # For regular FX graph, SymInt arg should be a fx.Node with
752
+ # self.is_sym_int_arg(arg) being true
753
+ return Argument.create(as_sym_int=SymIntArgument.create(as_name=str(arg)))
754
+ elif isinstance(arg, bool):
755
+ return Argument.create(as_bool=arg)
756
+ elif isinstance(arg, str):
757
+ return Argument.create(as_string=arg)
758
+ elif isinstance(arg, int):
759
+ return Argument.create(as_int=arg)
760
+ elif isinstance(arg, float):
761
+ return Argument.create(as_float=arg)
762
+ elif arg is None:
763
+ return Argument.create(as_none=())
764
+ elif isinstance(arg, (list, tuple)):
765
+ if len(arg) == 0:
766
+ if arg_type is not None:
767
+ if isinstance(arg_type, torch.OptionalType):
768
+ arg_type = arg_type.getElementType() # type: ignore[assignment]
769
+ assert isinstance(arg_type, torch.ListType)
770
+ elem_type = arg_type.getElementType()
771
+ if isinstance(elem_type, torch.OptionalType):
772
+ elem_type = elem_type.getElementType()
773
+
774
+ if isinstance(elem_type, torch.BoolType):
775
+ return Argument.create(as_bools=[])
776
+ elif isinstance(elem_type, torch.IntType):
777
+ return Argument.create(as_ints=[])
778
+ elif isinstance(elem_type, torch.FloatType):
779
+ return Argument.create(as_floats=[])
780
+ elif isinstance(elem_type, torch.StringType):
781
+ return Argument.create(as_strings=[])
782
+ elif isinstance(elem_type, torch.TensorType):
783
+ return Argument.create(as_tensors=[])
784
+ else:
785
+ # I believe empty symint lists default to ints, but
786
+ # please file an issue if this is not the case
787
+ raise SerializeError(f"Empty list with type {elem_type} nyi.")
788
+ else:
789
+ # We could serialize this by default to a tensor list. This
790
+ # is needed in the HOO case
791
+ log.warning(
792
+ "Unsure how to serialize the given empty list, "
793
+ "as we don't know what is the type of this argument. "
794
+ "Serializing it as a tensor list by default."
795
+ )
796
+ return Argument.create(as_tensors=[])
797
+
798
+ # Must check bool first, as bool is also treated as int
799
+ if all(isinstance(a, bool) for a in arg):
800
+ return Argument.create(as_bools=list(arg))
801
+ elif all(isinstance(a, int) for a in arg):
802
+ return Argument.create(as_ints=list(arg))
803
+ elif all(isinstance(a, float) for a in arg):
804
+ return Argument.create(as_floats=list(arg))
805
+ elif all(isinstance(a, str) for a in arg):
806
+ return Argument.create(as_strings=list(arg))
807
+ elif all(isinstance(a, torch.SymInt) for a in arg):
808
+ # This is a special branch for handling SymInt args in inductor's
809
+ # ExternalFallbackNode.
810
+ # For regular FX graph, SymInt arg should be a fx.Node with
811
+ # self.is_sym_int_arg(arg) being true
812
+ return Argument.create(
813
+ as_sym_ints=[SymIntArgument.create(as_name=str(a)) for a in arg]
814
+ )
815
+ elif all(self.is_sym_int_arg(a) for a in arg):
816
+ # list of sym_ints
817
+ values = []
818
+ for a in arg:
819
+ if isinstance(a, torch.fx.Node):
820
+ values.append(SymIntArgument.create(as_name=a.name))
821
+ elif isinstance(a, int):
822
+ values.append(SymIntArgument.create(as_int=a))
823
+ return Argument.create(as_sym_ints=values)
824
+ elif all(self.is_sym_bool_arg(a) for a in arg):
825
+ # list of sym_bools
826
+ values = []
827
+ for a in arg:
828
+ if isinstance(a, torch.fx.Node):
829
+ values.append(SymBoolArgument.create(as_name=a.name))
830
+ elif isinstance(a, bool):
831
+ values.append(SymBoolArgument.create(as_bool=a))
832
+ return Argument.create(as_sym_bools=values)
833
+ elif all(isinstance(a, torch.fx.Node) for a in arg):
834
+ # list of tensors
835
+ arguments = []
836
+ for a in arg:
837
+ if a.op == "get_attr":
838
+ raise SerializeError(
839
+ "getattr nodes containing tensors should not appear in the graph"
840
+ )
841
+ arguments.append(TensorArgument(name=a.name))
842
+ return Argument.create(as_tensors=arguments)
843
+ elif all(isinstance(a, (torch.fx.Node, type(None))) for a in arg):
844
+ # list of optional tensors
845
+ def serialize_optional_tensor_args(a):
846
+ if a is None:
847
+ return OptionalTensorArgument.create(as_none=())
848
+ elif isinstance(a, torch.fx.Node):
849
+ return OptionalTensorArgument.create(
850
+ as_tensor=TensorArgument(name=a.name)
851
+ )
852
+ else:
853
+ raise SerializeError(f"Unsupported list/tuple argument: {a}")
854
+
855
+ return Argument.create(
856
+ as_optional_tensors=list(map(serialize_optional_tensor_args, arg))
857
+ )
858
+ elif all(isinstance(a, inductor_tensor_buffers) for a in arg):
859
+ # list of inductor buffers
860
+ return Argument.create(
861
+ as_tensors=[TensorArgument(name=a.get_name()) for a in arg],
862
+ )
863
+ elif all(
864
+ isinstance(a, (*inductor_tensor_buffers, type(None))) for a in arg
865
+ ):
866
+ # list of inductor buffers as optional tensors
867
+ def serialize_optional_tensor_args(a):
868
+ if a is None:
869
+ return OptionalTensorArgument.create(as_none=())
870
+ elif isinstance(a, inductor_tensor_buffers):
871
+ return OptionalTensorArgument.create(
872
+ as_tensor=TensorArgument(name=a.get_name())
873
+ )
874
+ else:
875
+ raise SerializeError(f"Unsupported list/tuple argument: {a}")
876
+
877
+ return Argument.create(
878
+ as_optional_tensors=list(map(serialize_optional_tensor_args, arg))
879
+ )
880
+ else:
881
+ raise SerializeError(
882
+ f"Unsupported list/tuple argument type: {[type(a) for a in arg]}"
883
+ )
884
+ elif isinstance(arg, torch.dtype):
885
+ return Argument.create(as_scalar_type=_TORCH_TO_SERIALIZE_DTYPE[arg])
886
+ elif isinstance(arg, torch.device):
887
+ return Argument.create(as_device=Device(type=arg.type, index=arg.index))
888
+ elif isinstance(arg, torch.memory_format):
889
+ return Argument.create(
890
+ as_memory_format=_TORCH_TO_SERIALIZE_MEMORY_FORMAT[arg]
891
+ )
892
+ elif isinstance(arg, torch.layout):
893
+ return Argument.create(as_layout=_TORCH_TO_SERIALIZE_LAYOUT[arg])
894
+ elif isinstance(arg, torch._C.ScriptObject):
895
+ if not (
896
+ arg._has_method("__getstate__") # type: ignore[attr-defined]
897
+ and arg._has_method("__setstate__") # type: ignore[attr-defined]
898
+ ):
899
+ raise SerializeError(
900
+ f"Unable to serialize custom class {arg}. Please define "
901
+ "serialization methods via def_pickle()."
902
+ )
903
+ # Custom objects through torchind are serializable with pickle,
904
+ # through implementing the .def_pickle function. This should result
905
+ # in the object containing a __getstate__ and __setstate__
906
+ # serialize/deserialize function.
907
+ custom_obj_name = f"_custom_obj_{len(self.custom_objs)}"
908
+ self.custom_objs[custom_obj_name] = arg
909
+ class_fqn = arg._type().qualified_name() # type: ignore[attr-defined]
910
+ return Argument.create(
911
+ as_custom_obj=CustomObjArgument(custom_obj_name, class_fqn)
912
+ )
913
+ elif isinstance(arg, torch._ops.OpOverload):
914
+ return Argument.create(as_operator=self.serialize_operator(arg))
915
+ else:
916
+ raise SerializeError(f"Unsupported argument type: {type(arg)}")
917
+
918
+ def serialize_tensor_output(self, name, meta_val) -> TensorArgument:
919
+ assert name not in self.graph_state.tensor_values
920
+ self.graph_state.tensor_values[name] = serialize_tensor_meta(meta_val)
921
+ return TensorArgument(name=name)
922
+
923
+ def serialize_sym_int_output(self, name, meta_val) -> SymIntArgument:
924
+ assert name not in self.graph_state.sym_int_values
925
+ self.graph_state.sym_int_values[name] = serialize_sym_int(meta_val)
926
+ return SymIntArgument.create(as_name=name)
927
+
928
+ def serialize_sym_bool_output(self, name, meta_val) -> SymIntArgument:
929
+ assert name not in self.graph_state.sym_bool_values
930
+ self.graph_state.sym_bool_values[name] = serialize_sym_bool(meta_val)
931
+ return SymBoolArgument.create(as_name=name)
932
+
933
+ def serialize_input_spec(self, spec: ep.InputSpec) -> InputSpec:
934
+ if spec.kind == ep.InputKind.USER_INPUT:
935
+ if isinstance(spec.arg, ep.ConstantArgument):
936
+ if isinstance(spec.arg.value, int):
937
+ constant_spec = ConstantValue.create(as_int=spec.arg.value)
938
+ elif isinstance(spec.arg.value, bool):
939
+ constant_spec = ConstantValue.create(as_bool=spec.arg.value)
940
+ elif isinstance(spec.arg.value, str):
941
+ constant_spec = ConstantValue.create(as_string=spec.arg.value)
942
+ elif isinstance(spec.arg.value, float):
943
+ constant_spec = ConstantValue.create(as_float=spec.arg.value)
944
+ elif spec.arg.value is None:
945
+ constant_spec = ConstantValue.create(as_none=())
946
+ else:
947
+ raise SerializeError(
948
+ f"Unhandled constant input {spec.arg.value} to serialize"
949
+ )
950
+ return InputSpec.create(
951
+ constant_input=ConstantInputSpec(
952
+ name=spec.arg.name, value=constant_spec
953
+ )
954
+ )
955
+ else:
956
+ return InputSpec.create(
957
+ user_input=UserInputSpec(arg=self.serialize_argument_spec(spec.arg))
958
+ )
959
+ elif spec.kind == ep.InputKind.PARAMETER:
960
+ assert spec.target is not None
961
+ assert isinstance(spec.arg, ep.TensorArgument)
962
+ return InputSpec.create(
963
+ parameter=InputToParameterSpec(
964
+ arg=TensorArgument(name=spec.arg.name),
965
+ parameter_name=spec.target,
966
+ )
967
+ )
968
+ elif spec.kind == ep.InputKind.BUFFER:
969
+ assert spec.target is not None
970
+ assert isinstance(spec.arg, ep.TensorArgument)
971
+ assert spec.persistent is not None
972
+ return InputSpec.create(
973
+ buffer=InputToBufferSpec(
974
+ arg=TensorArgument(name=spec.arg.name),
975
+ buffer_name=spec.target,
976
+ persistent=spec.persistent,
977
+ )
978
+ )
979
+ elif spec.kind == ep.InputKind.CONSTANT_TENSOR:
980
+ assert spec.target is not None
981
+ assert isinstance(spec.arg, ep.TensorArgument)
982
+ return InputSpec.create(
983
+ tensor_constant=InputToTensorConstantSpec(
984
+ arg=TensorArgument(name=spec.arg.name),
985
+ tensor_constant_name=spec.target,
986
+ )
987
+ )
988
+ elif spec.kind == ep.InputKind.CUSTOM_OBJ:
989
+ assert spec.target is not None
990
+ assert isinstance(spec.arg, ep.CustomObjArgument)
991
+ return InputSpec.create(
992
+ custom_obj=InputToCustomObjSpec(
993
+ arg=CustomObjArgument(
994
+ name=spec.arg.name, class_fqn=spec.arg.class_fqn
995
+ ),
996
+ custom_obj_name=spec.target,
997
+ )
998
+ )
999
+ elif spec.kind == ep.InputKind.TOKEN:
1000
+ assert isinstance(spec.arg, ep.TokenArgument)
1001
+ return InputSpec.create(
1002
+ token=InputTokenSpec(
1003
+ arg=TokenArgument(name=spec.arg.name),
1004
+ )
1005
+ )
1006
+ else:
1007
+ raise AssertionError(f"Unknown argument kind: {spec}")
1008
+
1009
+ def serialize_output_spec(self, spec: ep.OutputSpec) -> OutputSpec:
1010
+ if spec.kind == ep.OutputKind.USER_OUTPUT:
1011
+ return OutputSpec.create(
1012
+ user_output=UserOutputSpec(arg=self.serialize_argument_spec(spec.arg))
1013
+ )
1014
+ elif spec.kind == ep.OutputKind.LOSS_OUTPUT:
1015
+ assert isinstance(spec.arg, ep.TensorArgument)
1016
+ return OutputSpec.create(
1017
+ loss_output=LossOutputSpec(arg=TensorArgument(name=spec.arg.name))
1018
+ )
1019
+ elif spec.kind == ep.OutputKind.BUFFER_MUTATION:
1020
+ assert spec.target is not None
1021
+ assert isinstance(spec.arg, ep.TensorArgument)
1022
+ return OutputSpec.create(
1023
+ buffer_mutation=BufferMutationSpec(
1024
+ arg=TensorArgument(name=spec.arg.name),
1025
+ buffer_name=spec.target,
1026
+ )
1027
+ )
1028
+ elif spec.kind == ep.OutputKind.GRADIENT_TO_PARAMETER:
1029
+ assert spec.target is not None
1030
+ assert isinstance(spec.arg, ep.TensorArgument)
1031
+ return OutputSpec.create(
1032
+ gradient_to_parameter=GradientToParameterSpec(
1033
+ arg=TensorArgument(name=spec.arg.name),
1034
+ parameter_name=spec.target,
1035
+ )
1036
+ )
1037
+ elif spec.kind == ep.OutputKind.GRADIENT_TO_USER_INPUT:
1038
+ assert spec.target is not None
1039
+ assert isinstance(spec.arg, ep.TensorArgument)
1040
+ return OutputSpec.create(
1041
+ gradient_to_user_input=GradientToUserInputSpec(
1042
+ arg=TensorArgument(name=spec.arg.name),
1043
+ user_input_name=spec.target,
1044
+ )
1045
+ )
1046
+ elif spec.kind == ep.OutputKind.USER_INPUT_MUTATION:
1047
+ assert spec.target is not None
1048
+ assert isinstance(spec.arg, ep.TensorArgument)
1049
+ return OutputSpec.create(
1050
+ user_input_mutation=UserInputMutationSpec(
1051
+ arg=TensorArgument(name=spec.arg.name),
1052
+ user_input_name=spec.target,
1053
+ )
1054
+ )
1055
+ elif spec.kind == ep.OutputKind.TOKEN:
1056
+ assert isinstance(spec.arg, ep.TokenArgument)
1057
+ return OutputSpec.create(
1058
+ token=OutputTokenSpec(
1059
+ arg=TokenArgument(name=spec.arg.name),
1060
+ )
1061
+ )
1062
+ else:
1063
+ raise AssertionError(f"Unknown argument kind: {spec}")
1064
+
1065
+ def serialize_signature(self, sig: ep.ExportGraphSignature) -> GraphSignature:
1066
+ return GraphSignature(
1067
+ input_specs=[self.serialize_input_spec(s) for s in sig.input_specs],
1068
+ output_specs=[self.serialize_output_spec(s) for s in sig.output_specs],
1069
+ )
1070
+
1071
+ def serialize_argument_spec(self, x: ep.ArgumentSpec) -> Argument:
1072
+ if isinstance(x, ep.TensorArgument):
1073
+ return Argument.create(as_tensor=TensorArgument(name=x.name))
1074
+ elif isinstance(x, ep.SymIntArgument):
1075
+ return Argument.create(as_sym_int=SymIntArgument.create(as_name=x.name))
1076
+ elif isinstance(x, ep.ConstantArgument):
1077
+ return self.serialize_input(x.value)
1078
+ elif isinstance(x, ep.CustomObjArgument):
1079
+ return Argument.create(
1080
+ as_custom_obj=CustomObjArgument(name=x.name, class_fqn=x.class_fqn)
1081
+ )
1082
+ else:
1083
+ raise AssertionError("TODO")
1084
+
1085
+ def serialize_module_call_signature(
1086
+ self, module_call_signature: ep.ModuleCallSignature
1087
+ ) -> ModuleCallSignature:
1088
+ return ModuleCallSignature(
1089
+ inputs=[
1090
+ self.serialize_argument_spec(x) for x in module_call_signature.inputs
1091
+ ],
1092
+ outputs=[
1093
+ self.serialize_argument_spec(x) for x in module_call_signature.outputs
1094
+ ],
1095
+ in_spec=treespec_dumps(module_call_signature.in_spec, TREESPEC_VERSION),
1096
+ out_spec=treespec_dumps(module_call_signature.out_spec, TREESPEC_VERSION),
1097
+ )
1098
+
1099
+ def serialize_module_call_graph(
1100
+ self, module_call_graph: List[ep.ModuleCallEntry]
1101
+ ) -> List[ModuleCallEntry]:
1102
+ return [
1103
+ ModuleCallEntry(
1104
+ fqn=entry.fqn,
1105
+ signature=(
1106
+ self.serialize_module_call_signature(entry.signature)
1107
+ if entry.signature
1108
+ else None
1109
+ ),
1110
+ )
1111
+ for entry in module_call_graph
1112
+ ]
1113
+
1114
+ def serialize_outputs(self, node: torch.fx.Node) -> List[Argument]:
1115
+ """For a given node, return the dataclass representing its output values.
1116
+
1117
+ [NOTE: Multiple outputs] We handle aggregates differently than FX. For
1118
+ FX, it looks like:
1119
+
1120
+ x = call_function("multiple_return", ...)
1121
+ element0 = call_function(getitem, x, 0)
1122
+ foo = call_function("use_output", element0)
1123
+
1124
+ We do not want the intermediate `getitem` call, so our serialized thing looks like:
1125
+
1126
+ element0, element1, element2 = call_function("multiple_return", ...)
1127
+ foo = call_function("use_output", element0)
1128
+
1129
+ We want names to be consistent across these two schemes, so that we can
1130
+ mostly reuse the names coming from FX. This function computes a mapping from
1131
+ the FX representation to our representation, preserving the names.
1132
+ """
1133
+ assert node.op == "call_function" and isinstance(
1134
+ node.target, torch._ops.OpOverload
1135
+ )
1136
+
1137
+ assert isinstance(node.target, torch._ops.OpOverload)
1138
+ returns = node.target._schema.returns
1139
+
1140
+ if len(returns) == 0:
1141
+ return []
1142
+
1143
+ meta_val = node.meta["val"]
1144
+
1145
+ # Check single value return
1146
+ if _is_single_tensor_list_return(node.target):
1147
+ # e.g "-> Tensor[]"
1148
+ tensor_args = []
1149
+ for idx, meta in enumerate(meta_val):
1150
+ user_node = _output_node_at_index(node, idx)
1151
+ name = (
1152
+ user_node.name
1153
+ if user_node is not None
1154
+ else f"{node.name}_unused_{idx}"
1155
+ )
1156
+ tensor_args.append(self.serialize_tensor_output(name, meta))
1157
+ return [Argument.create(as_tensors=tensor_args)]
1158
+ elif len(returns) == 1:
1159
+ return [self.serialize_output(node.name, meta_val)]
1160
+
1161
+ # There are a two possibilities at this point:
1162
+ # - This operator returns a tuple of Tensors, e.g. "-> (Tensor, Tensor)"
1163
+ # - This operator returns a tuple of mixed of Tensor and Tensors, e.g. "-> (Tensor, Tensor[])"
1164
+ #
1165
+ # Either way, start by gathering a list of TensorArguments with the correct names.
1166
+ # For consistent naming with FX, consult the downstream `getitem` node and
1167
+ # make sure our outputs have the same name.
1168
+
1169
+ output_arguments = []
1170
+ for idx, (meta, return_schema) in enumerate(zip(meta_val, returns)):
1171
+ if meta is None:
1172
+ assert isinstance(
1173
+ return_schema.real_type, (torch.OptionalType, torch.TensorType)
1174
+ )
1175
+ # When the return type is annoated as Tensor type, the op can also return an
1176
+ # undefined Tensor which will be implicitly converted to None in Python.
1177
+ output_arguments.append(Argument.create(as_none=()))
1178
+ elif isinstance(meta, FakeTensor):
1179
+ assert isinstance(
1180
+ return_schema.real_type, (torch.OptionalType, torch.TensorType)
1181
+ )
1182
+ user_node = _output_node_at_index(node, idx)
1183
+ name = (
1184
+ user_node.name
1185
+ if user_node is not None
1186
+ else f"{node.name}_unused_{idx}"
1187
+ )
1188
+ output_arguments.append(self.serialize_output(name, meta))
1189
+ elif isinstance(meta, list):
1190
+ # for List[Tensor] return type
1191
+ assert isinstance(
1192
+ return_schema.real_type, torch.ListType
1193
+ ) and isinstance(
1194
+ return_schema.real_type.getElementType(), torch.TensorType
1195
+ )
1196
+ user_node = _output_node_at_index(node, idx)
1197
+ assert user_node is not None
1198
+
1199
+ args = []
1200
+ for i, m in enumerate(meta):
1201
+ if m is None:
1202
+ continue
1203
+ sub_user_node = _output_node_at_index(user_node, i)
1204
+ assert sub_user_node is not None, f"No user found at index {i}"
1205
+
1206
+ args.append(self.serialize_tensor_output(sub_user_node.name, m))
1207
+ output_arguments.append(Argument.create(as_tensors=args))
1208
+ elif isinstance(meta, (int, SymInt)):
1209
+ user_node = _output_node_at_index(node, idx)
1210
+ name = (
1211
+ user_node.name
1212
+ if user_node is not None
1213
+ else f"{node.name}_unused_{idx}"
1214
+ )
1215
+ output_arguments.append(self.serialize_output(name, meta))
1216
+ else:
1217
+ raise ValueError(
1218
+ f"Unhandled output type {type(meta)} from node {node.format_node()}"
1219
+ )
1220
+
1221
+ return output_arguments
1222
+
1223
+ def serialize_hoo_outputs(self, node: torch.fx.Node) -> List[Argument]:
1224
+ """
1225
+ For serializing HOO outputs since HOOs do not have a schema.
1226
+ """
1227
+ meta_val = node.meta["val"]
1228
+
1229
+ if isinstance(meta_val, (tuple, list)):
1230
+ # Note: Since we don't have a schema, we just serialize all tuple
1231
+ # outputs to be a list of values. Even if the output is supposed to
1232
+ # be a tensor list (Tensor[]), we will serialize it to be a list of
1233
+ # tensors (Tensor, Tensor, Tensor). An exception is that if there's
1234
+ # a singleton tensor, we will serialize this to be a singleton
1235
+ # tensor list so that the deserializer knows to insert getitem nodes.
1236
+
1237
+ if len(meta_val) == 1:
1238
+ assert isinstance(meta_val[0], torch.Tensor)
1239
+ user_node = _output_node_at_index(node, 0)
1240
+ name = (
1241
+ user_node.name if user_node is not None else f"{node.name}_unused_0"
1242
+ )
1243
+ return [
1244
+ Argument.create(
1245
+ as_tensors=[self.serialize_tensor_output(name, meta_val[0])]
1246
+ )
1247
+ ]
1248
+
1249
+ outputs = []
1250
+ for i, element_meta_val in enumerate(meta_val):
1251
+ user_node = _output_node_at_index(node, i)
1252
+ if isinstance(element_meta_val, list):
1253
+ # e.g "-> Tensor[]"
1254
+ assert user_node is not None
1255
+
1256
+ tensors = []
1257
+ for j, m in enumerate(element_meta_val):
1258
+ if not isinstance(m, torch.Tensor):
1259
+ raise SerializeError(
1260
+ f"Serialize list output with type {type(m)} nyi"
1261
+ )
1262
+
1263
+ sub_user_node = _output_node_at_index(user_node, j)
1264
+ name = (
1265
+ sub_user_node.name
1266
+ if sub_user_node is not None
1267
+ else f"{user_node.name}_unused_{j}"
1268
+ )
1269
+ tensors.append(self.serialize_tensor_output(name, m))
1270
+ outputs.append(Argument.create(as_tensors=tensors))
1271
+
1272
+ else:
1273
+ name = (
1274
+ user_node.name
1275
+ if user_node is not None
1276
+ else f"{node.name}_unused_{i}"
1277
+ )
1278
+
1279
+ outputs.append(self.serialize_output(name, element_meta_val))
1280
+
1281
+ return outputs
1282
+ else:
1283
+ return [self.serialize_output(node.name, meta_val)]
1284
+
1285
+ def serialize_output(self, name: str, meta_val: Any) -> Argument:
1286
+ # Check single value return
1287
+ if meta_val is None:
1288
+ return Argument.create(as_none=())
1289
+ if isinstance(meta_val, torch.Tensor):
1290
+ # e.g "-> Tensor"
1291
+ return Argument.create(
1292
+ as_tensor=self.serialize_tensor_output(name, meta_val)
1293
+ )
1294
+ elif isinstance(meta_val, (int, torch.SymInt)):
1295
+ # e.g "-> SymInt"
1296
+ return Argument.create(
1297
+ as_sym_int=self.serialize_sym_int_output(name, meta_val)
1298
+ )
1299
+ elif isinstance(meta_val, torch.SymBool):
1300
+ # e.g "-> SymBool"
1301
+ return Argument.create(
1302
+ as_sym_bool=self.serialize_sym_bool_output(name, meta_val)
1303
+ )
1304
+
1305
+ # list outputs should've been handled earlier
1306
+ raise SerializeError(f"Unable to serialize output {meta_val}")
1307
+
1308
+ def _handle_getitem_users(self, node: torch.fx.Node) -> List[TensorArgument]:
1309
+ meta_val = node.meta["val"]
1310
+
1311
+ idx_to_name = {}
1312
+ for user in node.users:
1313
+ assert (
1314
+ user.target is operator.getitem
1315
+ ), f"User node {user} of {node} is incorrect"
1316
+ idx_to_name[user.args[1]] = user.name
1317
+
1318
+ for idx, _ in enumerate(meta_val):
1319
+ # FX does not emit a getitem node for any outputs that are unused.
1320
+ # However, we need a name for them so that the number of outputs will
1321
+ # correctly match the schema. Just assign a dummy name.
1322
+ if idx not in idx_to_name:
1323
+ idx_to_name[idx] = f"{node.name}_unused_{idx}"
1324
+
1325
+ arg_list = []
1326
+ for i, element_meta_val in enumerate(meta_val):
1327
+ arg_list.append(
1328
+ self.serialize_tensor_output(idx_to_name[i], element_meta_val)
1329
+ )
1330
+
1331
+ return arg_list
1332
+
1333
+ def serialize_graph(self, graph_module: torch.fx.GraphModule) -> Graph:
1334
+ assert isinstance(graph_module, torch.fx.GraphModule)
1335
+ for node in graph_module.graph.nodes:
1336
+ try:
1337
+ getattr(self, f"handle_{node.op}")(node)
1338
+ except Exception as e:
1339
+ raise SerializeError(
1340
+ f"Failed serializing node {node} in graph: {node.format_node()}"
1341
+ ) from e
1342
+
1343
+ return Graph(
1344
+ inputs=self.graph_state.inputs,
1345
+ nodes=self.graph_state.nodes,
1346
+ tensor_values=self.graph_state.tensor_values,
1347
+ sym_int_values=self.graph_state.sym_int_values,
1348
+ sym_bool_values=self.graph_state.sym_bool_values,
1349
+ custom_obj_values=self.graph_state.custom_obj_values,
1350
+ outputs=self.graph_state.outputs,
1351
+ is_single_tensor_return=self.graph_state.is_single_tensor_return,
1352
+ )
1353
+
1354
+ def serialize(self, graph_module: torch.fx.GraphModule) -> GraphModule:
1355
+ graph = self.serialize_graph(graph_module)
1356
+
1357
+ return GraphModule(
1358
+ graph=graph,
1359
+ signature=self.serialize_signature(self.graph_signature),
1360
+ module_call_graph=self.serialize_module_call_graph(self.module_call_graph),
1361
+ )
1362
+
1363
+
1364
+ class ExportedProgramSerializer:
1365
+ def __init__(self, opset_version: Optional[Dict[str, int]] = None):
1366
+ self.opset_version: Dict[str, int] = {}
1367
+ if opset_version:
1368
+ self.opset_version.update(opset_version)
1369
+ if "aten" not in self.opset_version:
1370
+ self.opset_version["aten"] = torch._C._get_max_operator_version()
1371
+
1372
+ def serialize(self, exported_program: ep.ExportedProgram) -> _SerializedProgram:
1373
+ """
1374
+ Args:
1375
+ exported_program: Exported Program to serialize
1376
+ """
1377
+ exported_program._validate()
1378
+
1379
+ gm_serializer = GraphModuleSerializer(
1380
+ exported_program.graph_signature, exported_program.module_call_graph
1381
+ )
1382
+ serialized_graph_module = gm_serializer.serialize(exported_program.graph_module)
1383
+ serialized_range_constraints = serialize_range_constraints(
1384
+ exported_program.range_constraints
1385
+ )
1386
+
1387
+ # TODO: Directly serialize exported_program.constants once
1388
+ # CustomClassHolders get stored in the ExportedProgram rather than in
1389
+ # the graph
1390
+ constants = {}
1391
+ for n, c in gm_serializer.custom_objs.items():
1392
+ constants[n] = c
1393
+ for n, t in exported_program.constants.items():
1394
+ assert n not in constants
1395
+ constants[n] = t
1396
+
1397
+ additional_kwargs = {}
1398
+ if hasattr(exported_program, "verifiers"):
1399
+ additional_kwargs["verifiers"] = [
1400
+ v.dialect for v in exported_program.verifiers
1401
+ ]
1402
+ elif hasattr(exported_program, "dialect"):
1403
+ additional_kwargs["dialect"] = exported_program.dialect
1404
+ serialized_ep = ExportedProgram(
1405
+ graph_module=serialized_graph_module,
1406
+ opset_version=self.opset_version,
1407
+ range_constraints=serialized_range_constraints,
1408
+ schema_version=SchemaVersion(
1409
+ major=SCHEMA_VERSION[0],
1410
+ minor=SCHEMA_VERSION[1],
1411
+ ),
1412
+ **additional_kwargs,
1413
+ )
1414
+
1415
+ # Test canonical form is well defined.
1416
+ canonicalize(serialized_ep)
1417
+
1418
+ return _SerializedProgram(
1419
+ serialized_ep,
1420
+ serialize_torch_artifact(exported_program.state_dict),
1421
+ serialize_torch_artifact(constants),
1422
+ serialize_torch_artifact(exported_program.example_inputs),
1423
+ )
1424
+
1425
+
1426
+ class GraphModuleDeserializer:
1427
+ @dataclasses.dataclass
1428
+ class Result:
1429
+ graph_module: torch.fx.GraphModule
1430
+ signature: ep.ExportGraphSignature
1431
+ module_call_graph: List[ep.ModuleCallEntry]
1432
+ names_to_symbols: Dict[str, sympy.Symbol]
1433
+ state_dict: Dict[str, Union[torch.Tensor, torch.nn.Parameter]]
1434
+ constants: Dict[str, Union[torch.Tensor, torch.ScriptObject]]
1435
+ example_inputs: Optional[Tuple[Tuple[torch.Tensor, ...], Dict[str, Any]]]
1436
+
1437
+ def __init__(self):
1438
+ self.serialized_name_to_node: Dict[str, torch.fx.Node] = {}
1439
+ self.serialized_name_to_meta: Dict[str, MetaType] = {}
1440
+ self.graph = torch.fx.Graph()
1441
+ self.module = torch.nn.Module()
1442
+
1443
+ @contextmanager
1444
+ def save_graph_module(self) -> Iterator[None]:
1445
+ saved = (
1446
+ self.graph,
1447
+ self.module,
1448
+ self.serialized_name_to_node,
1449
+ self.serialized_name_to_meta,
1450
+ )
1451
+ self.graph = torch.fx.Graph()
1452
+ self.module = torch.nn.Module()
1453
+ self.serialized_name_to_node = {}
1454
+ self.serialized_name_to_meta = {}
1455
+ try:
1456
+ yield
1457
+ finally:
1458
+ (
1459
+ self.graph,
1460
+ self.module,
1461
+ self.serialized_name_to_node,
1462
+ self.serialized_name_to_meta,
1463
+ ) = saved
1464
+
1465
+ def deserialize_operator(self, serialized_target: str):
1466
+ if serialized_target.startswith(
1467
+ "_operator"
1468
+ ): # TODO(zhxchen17) Follow up on this.
1469
+ module = operator
1470
+ serialized_target_names = serialized_target.split(".")[1:]
1471
+ elif serialized_target.startswith("torch"):
1472
+ module = torch # type: ignore[misc]
1473
+ serialized_target_names = serialized_target.split(".")[1:]
1474
+ else: # TODO(zhxchen17) Don't catch all here.
1475
+ return serialized_target
1476
+
1477
+ target = module
1478
+ for name in serialized_target_names:
1479
+ if not hasattr(target, name):
1480
+ return serialized_target
1481
+ else:
1482
+ target = getattr(target, name)
1483
+ return target
1484
+
1485
+ def deserialize_sym_int(self, s: SymInt) -> Union[int, torch.SymInt]:
1486
+ val = s.value
1487
+ if s.type == "as_expr":
1488
+ if val.hint is None:
1489
+ hint = None
1490
+ else:
1491
+ assert val.hint.type == "as_int"
1492
+ hint = val.hint.value
1493
+
1494
+ if val.expr_str in self.symbol_name_to_symbol:
1495
+ sym = self.symbol_name_to_symbol[val.expr_str]
1496
+ if (
1497
+ isinstance(sym, sympy.Symbol)
1498
+ and sym not in self.shape_env.var_to_val
1499
+ ):
1500
+ if hint is not None:
1501
+ self.shape_env.add_var_to_val(sym, hint)
1502
+ else:
1503
+ sym = sympy.sympify(val.expr_str, locals=self.symbol_name_to_symbol)
1504
+ # NOTE(avik): Assumptions on symbols are not explicitly serialized.
1505
+ # This seems dangerous: it might cause unknown differences in shape env behavior
1506
+ # on deserialization? Probably deserves a follow-up.
1507
+
1508
+ # Here we force symbols corresponding to SymInts to be at least integers.
1509
+ # Otherwise some expressions that the shape env would otherwise evaluate to False,
1510
+ # e.g., 2*s = 9, can have rational solutions, e.g., 9/2.
1511
+ sym = sym.subs(
1512
+ {s: sympy.Symbol(s.name, integer=True) for s in sym.free_symbols}
1513
+ )
1514
+ if isinstance(sym, sympy.Symbol):
1515
+ self.symbol_name_to_symbol[val.expr_str] = sym
1516
+ if hint is not None:
1517
+ self.shape_env.add_var_to_val(sym, hint)
1518
+
1519
+ if vr := self.symbol_name_to_range.get(val.expr_str):
1520
+ self.shape_env.constrain_symbol_range(
1521
+ sym,
1522
+ compiler_min=vr.lower, # type: ignore[arg-type]
1523
+ compiler_max=vr.upper, # type: ignore[arg-type]
1524
+ )
1525
+ else:
1526
+ # Placeholders, in particular, can have shapes as symbolic expressions.
1527
+ # We need to populate the shape env with the range constraints of their
1528
+ # free symbols, otherwise evaluating such expressions will error.
1529
+ self.symbol_name_to_symbol[val.expr_str] = sym
1530
+ free_symbols = sym.free_symbols
1531
+ for s in free_symbols:
1532
+ if s.name not in self.symbol_name_to_symbol:
1533
+ self.symbol_name_to_symbol[s.name] = s
1534
+ if vr := self.symbol_name_to_range.get(s.name):
1535
+ self.shape_env.constrain_symbol_range(
1536
+ s,
1537
+ compiler_min=vr.lower, # type: ignore[arg-type]
1538
+ compiler_max=vr.upper, # type: ignore[arg-type]
1539
+ )
1540
+
1541
+ return self.shape_env.create_symintnode(sym, hint=hint)
1542
+ elif s.type == "as_int":
1543
+ assert isinstance(val, int)
1544
+ return val
1545
+ else:
1546
+ raise SerializeError(
1547
+ f"SymInt has invalid field type {s.type} with value {s.value}"
1548
+ )
1549
+
1550
+ def deserialize_sym_bool(self, s: SymBool) -> Union[bool, torch.SymBool]:
1551
+ val = s.value
1552
+ if s.type == "as_expr":
1553
+ expr = sympy.sympify(val.expr_str, locals=self.symbol_name_to_symbol)
1554
+ return self.shape_env.create_symboolnode(expr)
1555
+ elif s.type == "as_bool":
1556
+ assert isinstance(val, bool)
1557
+ return val
1558
+ else:
1559
+ raise SerializeError(
1560
+ f"SymBool has invalid field type {s.type} with value {s.value}"
1561
+ )
1562
+
1563
+ def deserialize_tensor_meta(
1564
+ self,
1565
+ tensor_meta: TensorMeta,
1566
+ ) -> FakeTensor:
1567
+ with self.fake_tensor_mode:
1568
+ return cast(
1569
+ FakeTensor,
1570
+ torch.empty_strided(
1571
+ tuple(self.deserialize_sym_int(val) for val in tensor_meta.sizes), # type: ignore[misc]
1572
+ tuple(self.deserialize_sym_int(val) for val in tensor_meta.strides), # type: ignore[misc]
1573
+ device=deserialize_device(tensor_meta.device),
1574
+ dtype=_SERIALIZE_TO_TORCH_DTYPE[tensor_meta.dtype],
1575
+ ),
1576
+ )
1577
+
1578
+ def deserialize_script_obj_meta(
1579
+ self, script_obj_meta: CustomObjArgument
1580
+ ) -> ep.CustomObjArgument:
1581
+ return ep.CustomObjArgument(
1582
+ name=script_obj_meta.name,
1583
+ class_fqn=script_obj_meta.class_fqn,
1584
+ )
1585
+
1586
+ def deserialize_graph_output(self, output) -> Optional[Union[torch.fx.Node, int]]:
1587
+ if output.type == "as_tensor":
1588
+ return self.serialized_name_to_node[output.as_tensor.name]
1589
+ elif output.type == "as_sym_int":
1590
+ return self.serialized_name_to_node[output.as_sym_int.as_name]
1591
+ elif output.type == "as_sym_bool":
1592
+ return self.serialized_name_to_node[output.as_sym_bool.as_name]
1593
+ elif output.type == "as_int":
1594
+ return output.as_int
1595
+ elif output.type == "as_none":
1596
+ return None
1597
+ else:
1598
+ raise SerializeError(f"Unable to deserialize output node {output}")
1599
+
1600
+ def deserialize_graph(self, serialized_graph: Graph) -> torch.fx.Graph:
1601
+ # Handle the tensor metas.
1602
+ for name, tensor_value in serialized_graph.tensor_values.items():
1603
+ meta_val = self.deserialize_tensor_meta(tensor_value)
1604
+ self.serialized_name_to_meta[name] = meta_val
1605
+
1606
+ for name, sym_int_value in serialized_graph.sym_int_values.items():
1607
+ self.serialized_name_to_meta[name] = self.deserialize_sym_int(sym_int_value)
1608
+
1609
+ for name, sym_bool_value in serialized_graph.sym_bool_values.items():
1610
+ self.serialized_name_to_meta[name] = self.deserialize_sym_bool(
1611
+ sym_bool_value
1612
+ )
1613
+
1614
+ for name, script_obj_meta in serialized_graph.custom_obj_values.items():
1615
+ self.serialized_name_to_meta[name] = self.deserialize_script_obj_meta(
1616
+ script_obj_meta
1617
+ )
1618
+
1619
+ # Inputs: convert to placeholder nodes in FX.
1620
+ for i, input_ in enumerate(serialized_graph.inputs):
1621
+ if input_.type in ("as_tensor", "as_sym_int", "as_custom_obj"):
1622
+ if input_.type == "as_sym_int":
1623
+ node_name = input_.value.as_name
1624
+ else:
1625
+ node_name = input_.value.name
1626
+ placeholder_node = self.graph.placeholder(node_name)
1627
+ # FX might declare a name illegal (e.g. some nn.Modules use "input" as forward() arguments)
1628
+ # we will overwrite it
1629
+ placeholder_node.name = node_name
1630
+ self.sync_fx_node(node_name, placeholder_node)
1631
+ elif input_.type in (
1632
+ "as_int",
1633
+ "as_float",
1634
+ "as_bool",
1635
+ "as_none",
1636
+ "as_string",
1637
+ ):
1638
+ node_name = self.signature.input_specs[i].arg.name
1639
+ placeholder_node = self.graph.placeholder(node_name)
1640
+ placeholder_node.meta["val"] = self.deserialize_input(input_)
1641
+ else:
1642
+ raise SerializeError(f"Invalid input type {input_}")
1643
+
1644
+ # Nodes: convert to call_function nodes.
1645
+ for serialized_node in serialized_graph.nodes:
1646
+ try:
1647
+ target = self.deserialize_operator(serialized_node.target)
1648
+ self.deserialize_node(serialized_node, target)
1649
+
1650
+ except Exception as e:
1651
+ raise SerializeError(
1652
+ f"Failed deserializing node {serialized_node}"
1653
+ ) from e
1654
+
1655
+ # Outputs: convert to a single `output` node.
1656
+ outputs = []
1657
+ for output in serialized_graph.outputs:
1658
+ outputs.append(self.deserialize_graph_output(output))
1659
+
1660
+ if serialized_graph.is_single_tensor_return:
1661
+ assert len(outputs) == 1
1662
+ outputs = outputs[0] # type: ignore[assignment]
1663
+ else:
1664
+ outputs = tuple(outputs) # type: ignore[assignment]
1665
+
1666
+ output_node = self.graph.output(outputs)
1667
+
1668
+ if serialized_graph.is_single_tensor_return:
1669
+ output_node.meta["val"] = output_node.args[0].meta["val"]
1670
+ else:
1671
+ output_node.meta["val"] = tuple(
1672
+ arg.meta["val"] if isinstance(arg, torch.fx.Node) else arg
1673
+ for arg in output_node.args[0]
1674
+ )
1675
+
1676
+ return self.graph
1677
+
1678
+ def deserialize_node(self, serialized_node: Node, target: Callable) -> None:
1679
+ if target in _SYM_BOOL_OPS or target in _SYM_INT_OPS:
1680
+ name = serialized_node.name
1681
+ args = self.deserialize_sym_op_inputs(serialized_node.inputs)
1682
+
1683
+ fx_node = self.graph.create_node("call_function", target, args, {}, name)
1684
+ self.deserialize_sym_op_outputs(serialized_node, fx_node)
1685
+
1686
+ elif isinstance(target, torch._ops.HigherOrderOperator):
1687
+ args, kwargs = self.deserialize_hoo_inputs(serialized_node.inputs)
1688
+ # If HOP returns a single tensor, name the
1689
+ # newly-created node after it. This ensures that these tensor values
1690
+ # have names that are consistent with serialized.
1691
+ #
1692
+ # HOPs don't have schema yet, just check the output lengths and as_tensor attribute
1693
+ name = serialized_node.name
1694
+ fx_node = self.graph.create_node(
1695
+ "call_function", target, args, kwargs, name
1696
+ )
1697
+ self.deserialize_outputs(serialized_node, fx_node)
1698
+ fx_node.meta.update(self.deserialize_metadata(serialized_node.metadata))
1699
+
1700
+ elif isinstance(target, torch._ops.OpOverload):
1701
+ # For convenience: if this node returns a single tensor, name the
1702
+ # newly-created node after it. This ensures that these tensor values
1703
+ # have names that are consistent with serialized.
1704
+
1705
+ name = serialized_node.name
1706
+
1707
+ args, kwargs = self.deserialize_inputs(target, serialized_node)
1708
+ fx_node = self.graph.create_node(
1709
+ "call_function", target, args, kwargs, name
1710
+ )
1711
+ self.deserialize_outputs(serialized_node, fx_node)
1712
+ else:
1713
+ raise SerializeError(
1714
+ f"Unsupported target type for node {serialized_node}: {target}"
1715
+ )
1716
+
1717
+ fx_node.meta.update(self.deserialize_metadata(serialized_node.metadata))
1718
+ if (
1719
+ fx_node.op not in ["placeholder", "output"]
1720
+ and "nn_module_stack" not in fx_node.meta
1721
+ ):
1722
+ fx_node.meta["nn_module_stack"] = (
1723
+ {}
1724
+ ) # serialization throws away empty dicts
1725
+
1726
+ def deserialize_input_spec(self, i: InputSpec) -> ep.InputSpec:
1727
+ if i.type == "user_input":
1728
+ return ep.InputSpec(
1729
+ kind=ep.InputKind.USER_INPUT,
1730
+ arg=self.deserialize_argument_spec(i.user_input.arg),
1731
+ target=None,
1732
+ )
1733
+ elif i.type == "parameter":
1734
+ return ep.InputSpec(
1735
+ kind=ep.InputKind.PARAMETER,
1736
+ arg=ep.TensorArgument(name=i.parameter.arg.name),
1737
+ target=i.parameter.parameter_name,
1738
+ )
1739
+ elif i.type == "buffer":
1740
+ return ep.InputSpec(
1741
+ kind=ep.InputKind.BUFFER,
1742
+ arg=ep.TensorArgument(name=i.buffer.arg.name),
1743
+ target=i.buffer.buffer_name,
1744
+ persistent=i.buffer.persistent,
1745
+ )
1746
+ elif i.type == "tensor_constant":
1747
+ return ep.InputSpec(
1748
+ kind=ep.InputKind.CONSTANT_TENSOR,
1749
+ arg=ep.TensorArgument(name=i.tensor_constant.arg.name),
1750
+ target=i.tensor_constant.tensor_constant_name,
1751
+ )
1752
+ elif i.type == "custom_obj":
1753
+ return ep.InputSpec(
1754
+ kind=ep.InputKind.CUSTOM_OBJ,
1755
+ arg=ep.CustomObjArgument(
1756
+ name=i.custom_obj.arg.name, class_fqn=i.custom_obj.arg.class_fqn
1757
+ ),
1758
+ target=i.custom_obj.custom_obj_name,
1759
+ )
1760
+ elif i.type == "token":
1761
+ return ep.InputSpec(
1762
+ kind=ep.InputKind.TOKEN,
1763
+ arg=ep.TokenArgument(name=i.token.arg.name),
1764
+ target=None,
1765
+ )
1766
+ elif i.type == "constant_input":
1767
+ return ep.InputSpec(
1768
+ kind=ep.InputKind.USER_INPUT,
1769
+ arg=ep.ConstantArgument(
1770
+ name=i.constant_input.name,
1771
+ value=self.deserialize_constant_input(i.constant_input.value),
1772
+ ),
1773
+ target=None,
1774
+ )
1775
+ else:
1776
+ raise AssertionError(f"Unknown input spec {i}")
1777
+
1778
+ def deserialize_output_spec(self, o: OutputSpec) -> ep.OutputSpec:
1779
+ if o.type == "user_output":
1780
+ return ep.OutputSpec(
1781
+ kind=ep.OutputKind.USER_OUTPUT,
1782
+ arg=self.deserialize_argument_spec(o.user_output.arg),
1783
+ target=None,
1784
+ )
1785
+ elif o.type == "loss_output":
1786
+ return ep.OutputSpec(
1787
+ kind=ep.OutputKind.LOSS_OUTPUT,
1788
+ arg=ep.TensorArgument(name=o.loss_output.arg.name),
1789
+ target=None,
1790
+ )
1791
+ elif o.type == "buffer_mutation":
1792
+ return ep.OutputSpec(
1793
+ kind=ep.OutputKind.BUFFER_MUTATION,
1794
+ arg=ep.TensorArgument(name=o.buffer_mutation.arg.name),
1795
+ target=o.buffer_mutation.buffer_name,
1796
+ )
1797
+ elif o.type == "gradient_to_parameter":
1798
+ return ep.OutputSpec(
1799
+ kind=ep.OutputKind.GRADIENT_TO_PARAMETER,
1800
+ arg=ep.TensorArgument(name=o.gradient_to_parameter.arg.name),
1801
+ target=o.gradient_to_parameter.parameter_name,
1802
+ )
1803
+ elif o.type == "gradient_to_user_input":
1804
+ return ep.OutputSpec(
1805
+ kind=ep.OutputKind.GRADIENT_TO_USER_INPUT,
1806
+ arg=ep.TensorArgument(name=o.gradient_to_user_input.arg.name),
1807
+ target=o.gradient_to_user_input.user_input_name,
1808
+ )
1809
+ elif o.type == "user_input_mutation":
1810
+ return ep.OutputSpec(
1811
+ kind=ep.OutputKind.USER_INPUT_MUTATION,
1812
+ arg=ep.TensorArgument(name=o.user_input_mutation.arg.name),
1813
+ target=o.user_input_mutation.user_input_name,
1814
+ )
1815
+ elif o.type == "token":
1816
+ return ep.OutputSpec(
1817
+ kind=ep.OutputKind.TOKEN,
1818
+ arg=ep.TokenArgument(name=o.token.arg.name),
1819
+ target=None,
1820
+ )
1821
+ else:
1822
+ raise AssertionError(f"Unknown output spec {o}")
1823
+
1824
+ def deserialize_signature(self, sig: GraphSignature) -> ep.ExportGraphSignature:
1825
+ return ep.ExportGraphSignature(
1826
+ input_specs=[self.deserialize_input_spec(i) for i in sig.input_specs],
1827
+ output_specs=[self.deserialize_output_spec(o) for o in sig.output_specs],
1828
+ )
1829
+
1830
+ def deserialize(
1831
+ self,
1832
+ serialized_graph_module: GraphModule,
1833
+ serialized_state_dict: Union[Dict[str, torch.Tensor], bytes],
1834
+ constants: Union[Dict[str, Any], bytes],
1835
+ example_inputs: Optional[
1836
+ Union[Tuple[Tuple[torch.Tensor, ...], Dict[str, Any]], bytes]
1837
+ ] = None,
1838
+ symbol_name_to_range: Optional[Dict[str, symbolic_shapes.ValueRanges]] = None,
1839
+ ) -> Result:
1840
+ global _CURRENT_DESERIALIZER
1841
+ current_deserializer_state = _CURRENT_DESERIALIZER.copy()
1842
+ _CURRENT_DESERIALIZER.append(self)
1843
+ try:
1844
+ self.shape_env = symbolic_shapes.ShapeEnv(assume_static_by_default=True)
1845
+ self.fake_tensor_mode = FakeTensorMode(
1846
+ allow_fallback_kernels=False,
1847
+ allow_non_fake_inputs=True,
1848
+ shape_env=self.shape_env,
1849
+ )
1850
+ self.symbol_name_to_symbol: Dict[str, sympy.Symbol] = {}
1851
+ self.constants = deserialize_torch_artifact(constants)
1852
+ self.signature = self.deserialize_signature(
1853
+ serialized_graph_module.signature
1854
+ )
1855
+
1856
+ # deserialization does analysis with checks on 0/1, so we create fake range constraints and
1857
+ # restore the original range constraints afterwards
1858
+ self.symbol_name_to_range = {}
1859
+ if symbol_name_to_range:
1860
+ for k, vr in symbol_name_to_range.items():
1861
+ if math.isinf(vr.lower) and vr.lower < 0:
1862
+ lower = -math.inf
1863
+ elif math.isinf(vr.lower):
1864
+ lower = math.inf
1865
+ else:
1866
+ lower = int(vr.lower)
1867
+
1868
+ if vr.upper >= 2: # max is >= 2, not sym bool range
1869
+ lower = max(2, lower)
1870
+ self.symbol_name_to_range[k] = symbolic_shapes.ValueRanges(
1871
+ _int_to_sympy_int(lower), vr.upper
1872
+ )
1873
+
1874
+ if example_inputs is not None and len(example_inputs) > 0:
1875
+ self.example_inputs = deserialize_torch_artifact(example_inputs)
1876
+ else:
1877
+ self.example_inputs = None
1878
+ self.deserialize_graph(serialized_graph_module.graph)
1879
+
1880
+ module_call_graph = self.deserialize_module_call_graph(
1881
+ serialized_graph_module.module_call_graph
1882
+ )
1883
+ return GraphModuleDeserializer.Result(
1884
+ graph_module=ep._create_graph_module_for_export(
1885
+ self.module, self.graph
1886
+ ),
1887
+ signature=self.signature,
1888
+ module_call_graph=module_call_graph,
1889
+ names_to_symbols=self.symbol_name_to_symbol,
1890
+ state_dict=deserialize_torch_artifact(serialized_state_dict),
1891
+ constants=self.constants,
1892
+ example_inputs=self.example_inputs,
1893
+ )
1894
+ finally:
1895
+ _CURRENT_DESERIALIZER.pop()
1896
+ assert current_deserializer_state == _CURRENT_DESERIALIZER
1897
+
1898
+ def sync_fx_node(self, name: str, fx_node: torch.fx.Node):
1899
+ if name in self.serialized_name_to_node:
1900
+ raise SerializeError(f"Node {name} has already been deserialized before.")
1901
+ self.serialized_name_to_node[name] = fx_node
1902
+ assert "val" not in fx_node.meta
1903
+ fx_node.meta["val"] = self.serialized_name_to_meta[name]
1904
+
1905
+ def deserialize_sym_op_inputs(self, inputs):
1906
+ return tuple(self.deserialize_input(input.arg) for input in inputs)
1907
+
1908
+ def deserialize_inputs(self, target: torch._ops.OpOverload, serialized_node: Node):
1909
+ schema_args = target._schema.arguments
1910
+ actual_args = {
1911
+ input.name: self.deserialize_input(input.arg)
1912
+ for input in serialized_node.inputs
1913
+ }
1914
+ args = []
1915
+ kwargs = {}
1916
+ for schema_arg in schema_args:
1917
+ is_positional = (
1918
+ not schema_arg.has_default_value() and not schema_arg.kwarg_only
1919
+ )
1920
+ if is_positional:
1921
+ args.append(actual_args[schema_arg.name])
1922
+ else:
1923
+ if schema_arg.name in actual_args:
1924
+ kwargs[schema_arg.name] = actual_args[schema_arg.name]
1925
+ return tuple(args), kwargs
1926
+
1927
+ def deserialize_hoo_inputs(self, inputs: List[NamedArgument]):
1928
+ """
1929
+ For deserializing HOO inputs since HOOs do not have a schema.
1930
+ """
1931
+ args = []
1932
+ kwargs = {}
1933
+ for input_ in inputs:
1934
+ if input_.name != "":
1935
+ kwargs[input_.name] = self.deserialize_input(input_.arg)
1936
+ else:
1937
+ args.append(self.deserialize_input(input_.arg))
1938
+ return (tuple(args), kwargs)
1939
+
1940
+ def deserialize_input(self, inp: Argument) -> Any:
1941
+ value = inp.value
1942
+ typ_ = inp.type
1943
+ if typ_ == "as_none":
1944
+ # None should converted as None, but is encoded as bool in serialized
1945
+ # Convert serialized object to torch equivalent
1946
+ return None
1947
+ elif typ_ == "as_tensor":
1948
+ return self.serialized_name_to_node[inp.as_tensor.name]
1949
+ elif typ_ == "as_scalar_type":
1950
+ return _SERIALIZE_TO_TORCH_DTYPE[inp.as_scalar_type]
1951
+ elif typ_ == "as_memory_format":
1952
+ return _SERIALIZE_TO_TORCH_MEMORY_FORMAT[inp.as_memory_format]
1953
+ elif typ_ == "as_layout":
1954
+ return _SERIALIZE_TO_TORCH_LAYOUT[inp.as_layout]
1955
+ elif typ_ == "as_graph":
1956
+ assert isinstance(value, GraphArgument)
1957
+ with self.save_graph_module():
1958
+ self.deserialize_graph(value.graph)
1959
+ submodule = ep._create_graph_module_for_export(self.module, self.graph)
1960
+ self.module.register_module(value.name, submodule)
1961
+ return self.graph.create_node(
1962
+ "get_attr",
1963
+ value.name,
1964
+ name=value.name,
1965
+ )
1966
+ elif typ_ == "as_device":
1967
+ return deserialize_device(inp.as_device)
1968
+ elif typ_ == "as_int":
1969
+ return inp.as_int
1970
+ elif typ_ == "as_float":
1971
+ return inp.as_float
1972
+ elif typ_ == "as_bool":
1973
+ return inp.as_bool
1974
+ elif typ_ == "as_string":
1975
+ return inp.as_string
1976
+ elif typ_ == "as_sym_int":
1977
+ return self.deserialize_sym_argument(inp.as_sym_int)
1978
+ elif typ_ == "as_sym_bool":
1979
+ return self.deserialize_sym_argument(inp.as_sym_bool)
1980
+ elif isinstance(value, list):
1981
+ if len(value) == 0:
1982
+ return []
1983
+ elif typ_ == "as_tensors":
1984
+ result = []
1985
+ for arg in value:
1986
+ result.append(self.serialized_name_to_node[arg.name])
1987
+ return result
1988
+ elif typ_ in ("as_ints", "as_floats", "as_bools", "as_strings"):
1989
+ # convert from serialized.python.types.List to python list
1990
+ return list(value)
1991
+ elif typ_ in ("as_sym_ints", "as_sym_bools"):
1992
+ return [self.deserialize_sym_argument(arg) for arg in value]
1993
+ elif typ_ == "as_optional_tensors":
1994
+
1995
+ def deserialize_optional_tensor_args(a):
1996
+ if a.type == "as_none":
1997
+ return None
1998
+ elif a.type == "as_tensor":
1999
+ return self.serialized_name_to_node[a.value.name]
2000
+ else:
2001
+ raise SerializeError(f"Unhandled argument {inp}")
2002
+
2003
+ return list(map(deserialize_optional_tensor_args, value))
2004
+ else:
2005
+ raise SerializeError(f"Unhandled argument {inp}")
2006
+ elif typ_ == "as_custom_obj":
2007
+ if inp.as_custom_obj.name in self.serialized_name_to_node:
2008
+ # Custom object has been lifted as an input
2009
+ return self.serialized_name_to_node[inp.as_custom_obj.name]
2010
+ return self.constants[inp.as_custom_obj.name]
2011
+ elif typ_ == "as_operator":
2012
+ return self.deserialize_operator(inp.as_operator)
2013
+ else:
2014
+ raise SerializeError(f"Unhandled argument {inp}")
2015
+
2016
+ def deserialize_constant_input(self, inp: ConstantValue) -> Any:
2017
+ if inp.type == "as_int":
2018
+ return int(inp.as_int)
2019
+ elif inp.type == "as_float":
2020
+ return float(inp.as_float)
2021
+ elif inp.type == "as_string":
2022
+ return str(inp.as_string)
2023
+ elif inp.type == "as_bool":
2024
+ return bool(inp.as_bool)
2025
+ elif inp.type == "as_none":
2026
+ return None
2027
+ else:
2028
+ raise SerializeError(f"Unhandled constant argument {inp} to deserialize")
2029
+
2030
+ def deserialize_sym_argument(self, sym_arg):
2031
+ if isinstance(sym_arg, SymIntArgument):
2032
+ if sym_arg.type == "as_int":
2033
+ return sym_arg.as_int
2034
+ elif sym_arg.type == "as_name":
2035
+ return self.serialized_name_to_node[sym_arg.as_name]
2036
+ elif isinstance(sym_arg, SymBoolArgument):
2037
+ if sym_arg.type == "as_bool":
2038
+ return sym_arg.as_bool
2039
+ elif sym_arg.type == "as_name":
2040
+ return self.serialized_name_to_node[sym_arg.as_name]
2041
+ raise SerializeError(f"Unknown symbolic argument type: {sym_arg}")
2042
+
2043
+ def deserialize_sym_op_outputs(self, serialized_node: Node, fx_node: torch.fx.Node):
2044
+ self.sync_fx_node(serialized_node.outputs[0].value.as_name, fx_node)
2045
+
2046
+ def deserialize_outputs(self, serialized_node: Node, fx_node: torch.fx.Node):
2047
+ # Check single value return
2048
+ if len(serialized_node.outputs) == 0:
2049
+ return
2050
+ if (
2051
+ len(serialized_node.outputs) == 1
2052
+ and serialized_node.outputs[0].type == "as_tensor"
2053
+ ):
2054
+ self.sync_fx_node(serialized_node.outputs[0].as_tensor.name, fx_node)
2055
+ return
2056
+ elif len(serialized_node.outputs) == 1 and isinstance(
2057
+ serialized_node.outputs[0].value, (SymIntArgument, SymBoolArgument)
2058
+ ):
2059
+ self.sync_fx_node(serialized_node.outputs[0].value.as_name, fx_node)
2060
+ return
2061
+
2062
+ self.deserialize_multiple_outputs(serialized_node, fx_node)
2063
+
2064
+ def deserialize_multiple_outputs(
2065
+ self, serialized_node: Node, fx_node: torch.fx.Node
2066
+ ) -> None:
2067
+ deserialized_metadata = self.deserialize_metadata(serialized_node.metadata)
2068
+
2069
+ def generate_getitem(
2070
+ meta_val,
2071
+ fx_node: torch.fx.Node,
2072
+ arg: Union[TensorArgument, SymIntArgument],
2073
+ idx: int,
2074
+ ):
2075
+ if isinstance(arg, TensorArgument):
2076
+ name = arg.name
2077
+ elif isinstance(arg, SymIntArgument):
2078
+ name = arg.as_name
2079
+ else:
2080
+ raise AssertionError(
2081
+ f"generate_getitem got unknown argument type {type(arg)}"
2082
+ )
2083
+ individual_output = self.graph.create_node(
2084
+ "call_function",
2085
+ operator.getitem,
2086
+ (fx_node, idx),
2087
+ name=name,
2088
+ )
2089
+ self.sync_fx_node(name, individual_output)
2090
+ meta_val.append(self.serialized_name_to_meta[name])
2091
+ # The derived `getitem` nodes should have the same stacktrace as the
2092
+ # original `fx_node`
2093
+ individual_output.meta.update(deserialized_metadata)
2094
+
2095
+ def generate_getitems(meta_val, fx_node: torch.fx.Node, args):
2096
+ for idx, arg in enumerate(args):
2097
+ if isinstance(arg, Argument):
2098
+ arg = arg.value
2099
+ if isinstance(arg, (TensorArgument, SymIntArgument)):
2100
+ generate_getitem(meta_val, fx_node, arg, idx)
2101
+ elif isinstance(arg, (list, tuple)):
2102
+ list_output = self.graph.create_node(
2103
+ "call_function",
2104
+ operator.getitem,
2105
+ (fx_node, idx),
2106
+ )
2107
+ meta_val.append([])
2108
+ generate_getitems(meta_val[-1], list_output, arg)
2109
+ list_output.meta.update(deserialized_metadata)
2110
+ list_output.meta["val"] = meta_val[-1]
2111
+ else:
2112
+ raise NotImplementedError(f"Unimplemented node output type: {arg}")
2113
+
2114
+ # Convert multiple return types to FX format.
2115
+ # In FX, each node only returns one value. So in order to represent
2116
+ # multiple return values, we have to emit a `getitem` node for each
2117
+ # return value.
2118
+ # This performs the inverse mapping of the `serialize_outputs` call in
2119
+ # serialization, see [NOTE: Multiple outputs]
2120
+ meta_val: List[Any] = []
2121
+ if len(serialized_node.outputs) == 1:
2122
+ assert isinstance(serialized_node.outputs[0].value, list)
2123
+ assert isinstance(serialized_node.outputs[0].value[0], TensorArgument)
2124
+ generate_getitems(meta_val, fx_node, serialized_node.outputs[0].as_tensors)
2125
+ else:
2126
+ generate_getitems(meta_val, fx_node, serialized_node.outputs)
2127
+
2128
+ # also update the metaval for `fx_node` to be a list(meta)
2129
+ fx_node.meta["val"] = tuple(meta_val)
2130
+ self.serialized_name_to_node[fx_node.name] = fx_node
2131
+
2132
+ def deserialize_metadata(self, metadata: Dict[str, str]) -> Dict[str, Any]:
2133
+ ret: Dict[str, Any] = {}
2134
+ if stack_trace := metadata.get("stack_trace"):
2135
+ ret["stack_trace"] = stack_trace
2136
+
2137
+ def deserialize_meta_func(serialized_target: str):
2138
+ module = None
2139
+ if serialized_target.startswith("torch.nn"):
2140
+ module = torch.nn
2141
+ serialized_target_names = serialized_target.split(".")[2:]
2142
+ elif serialized_target.startswith("torch"):
2143
+ module = torch
2144
+ serialized_target_names = serialized_target.split(".")[1:]
2145
+ else:
2146
+ return self.deserialize_operator(serialized_target)
2147
+
2148
+ target = module
2149
+ for name in serialized_target_names:
2150
+ if not hasattr(target, name):
2151
+ return serialized_target
2152
+ else:
2153
+ target = getattr(target, name)
2154
+ return target
2155
+
2156
+ if nn_module_stack_str := metadata.get("nn_module_stack"):
2157
+ # Originally serialized to "key,orig_path,type_str"
2158
+ def import_nn_module_stack(key, path, ty):
2159
+ return key, (path, ty)
2160
+
2161
+ # Helper function to split string by commas, accounting for nested parentheses/brackets
2162
+ def metadata_split(metadata):
2163
+ out = []
2164
+ start, depth = 0, 0
2165
+ for position, char in enumerate(metadata):
2166
+ if char in "[(":
2167
+ depth += 1
2168
+ elif char in ")]":
2169
+ depth -= 1
2170
+ if depth < 0:
2171
+ raise ValueError(f"Mismatched brackets in metadata: {metadata}")
2172
+ elif char == "," and depth == 0:
2173
+ out.append(metadata[start:position].strip())
2174
+ start = position + 1
2175
+ out.append(metadata[start:].strip())
2176
+ assert len(out) == 3
2177
+ return out
2178
+
2179
+ nn_module_stack = dict(
2180
+ import_nn_module_stack(*metadata_split(item))
2181
+ for item in nn_module_stack_str.split(ST_DELIMITER)
2182
+ )
2183
+ ret["nn_module_stack"] = nn_module_stack
2184
+
2185
+ if source_fn_st_str := metadata.get("source_fn_stack"):
2186
+ # Originally serializes to "fx_node_name,op_str"
2187
+ source_fn_st = []
2188
+ for source_fn_str in source_fn_st_str.split(ST_DELIMITER):
2189
+ name, target_str = source_fn_str.split(",")
2190
+ source_fn_st.append((name, deserialize_meta_func(target_str)))
2191
+ ret["source_fn_stack"] = source_fn_st
2192
+
2193
+ if torch_fn_str := metadata.get("torch_fn"):
2194
+ ret["torch_fn"] = tuple(torch_fn_str.split(ST_DELIMITER))
2195
+ return ret
2196
+
2197
+ def deserialize_argument_spec(self, x: Argument) -> ep.ArgumentSpec:
2198
+ if x.type == "as_tensor":
2199
+ return ep.TensorArgument(name=x.as_tensor.name)
2200
+ elif x.type == "as_sym_int":
2201
+ return ep.SymIntArgument(name=x.as_sym_int.as_name)
2202
+ elif x.type == "as_custom_obj":
2203
+ return ep.ConstantArgument(
2204
+ name=x.as_custom_obj.name, value=self.deserialize_input(x)
2205
+ )
2206
+ else:
2207
+ return ep.ConstantArgument(name="", value=self.deserialize_input(x))
2208
+
2209
+ def deserialize_module_call_signature(
2210
+ self, module_call_signature: ModuleCallSignature
2211
+ ) -> ep.ModuleCallSignature:
2212
+ return ep.ModuleCallSignature(
2213
+ inputs=[
2214
+ self.deserialize_argument_spec(x) for x in module_call_signature.inputs
2215
+ ],
2216
+ outputs=[
2217
+ self.deserialize_argument_spec(x) for x in module_call_signature.outputs
2218
+ ],
2219
+ in_spec=treespec_loads(module_call_signature.in_spec),
2220
+ out_spec=treespec_loads(module_call_signature.out_spec),
2221
+ )
2222
+
2223
+ def deserialize_module_call_graph(
2224
+ self, module_call_graph: List[ModuleCallEntry]
2225
+ ) -> List[ep.ModuleCallEntry]:
2226
+ return [
2227
+ ep.ModuleCallEntry(
2228
+ fqn=entry.fqn,
2229
+ signature=(
2230
+ self.deserialize_module_call_signature(entry.signature)
2231
+ if entry.signature
2232
+ else None
2233
+ ),
2234
+ )
2235
+ for entry in module_call_graph
2236
+ ]
2237
+
2238
+
2239
+ class ExportedProgramDeserializer:
2240
+ def __init__(self):
2241
+ pass
2242
+
2243
+ def deserialize_range_constraints(
2244
+ self,
2245
+ symbol_name_to_range: Dict[str, symbolic_shapes.ValueRanges],
2246
+ symbol_name_to_symbol: Dict[str, sympy.Symbol],
2247
+ ) -> Dict[sympy.Symbol, ValueRanges]:
2248
+ range_constraints = {}
2249
+ for k, v in symbol_name_to_range.items():
2250
+ if symbol := symbol_name_to_symbol.get(k):
2251
+ range_constraints[symbol] = v # type: ignore[arg-type]
2252
+ else:
2253
+ log.warning(f"Symbol {k} did not appear in the graph that was deserialized") # noqa: G004
2254
+ return range_constraints
2255
+
2256
+ def deserialize(
2257
+ self,
2258
+ exported_program: ExportedProgram,
2259
+ state_dict: Union[Dict[str, torch.Tensor], bytes],
2260
+ constants: Union[Dict[str, torch.Tensor], bytes],
2261
+ example_inputs: Optional[
2262
+ Union[Tuple[Tuple[torch.Tensor, ...], Dict[str, Any]], bytes]
2263
+ ] = None,
2264
+ ) -> ep.ExportedProgram:
2265
+ assert isinstance(exported_program, ExportedProgram)
2266
+ version = exported_program.schema_version
2267
+
2268
+ # TODO(zhxchen17) blocked on thrift schema refactor
2269
+ if version.major != SCHEMA_VERSION[0] and not (
2270
+ version.major == 0 and version.minor == 0
2271
+ ):
2272
+ raise SerializeError(
2273
+ f"Serialized schema version {exported_program.schema_version} "
2274
+ f"does not match our current schema version {SCHEMA_VERSION}."
2275
+ )
2276
+
2277
+ symbol_name_to_range = {
2278
+ k: symbolic_shapes.ValueRanges(
2279
+ _int_to_sympy_int(v.min_val), _int_to_sympy_int(v.max_val)
2280
+ )
2281
+ for k, v in exported_program.range_constraints.items()
2282
+ }
2283
+ res = GraphModuleDeserializer().deserialize(
2284
+ exported_program.graph_module,
2285
+ state_dict,
2286
+ constants,
2287
+ example_inputs,
2288
+ symbol_name_to_range,
2289
+ )
2290
+ range_constraints = self.deserialize_range_constraints(
2291
+ symbol_name_to_range,
2292
+ res.names_to_symbols,
2293
+ )
2294
+ exported_program = ep.ExportedProgram(
2295
+ root=res.graph_module,
2296
+ graph=res.graph_module.graph,
2297
+ graph_signature=res.signature,
2298
+ state_dict=res.state_dict, # type: ignore[arg-type]
2299
+ range_constraints=range_constraints,
2300
+ module_call_graph=res.module_call_graph,
2301
+ example_inputs=res.example_inputs,
2302
+ verifier=load_verifier(exported_program.dialect),
2303
+ constants=res.constants,
2304
+ )
2305
+ return exported_program
2306
+
2307
+
2308
+ class EnumEncoder(json.JSONEncoder):
2309
+ def default(self, obj):
2310
+ if isinstance(obj, Enum):
2311
+ return obj.value
2312
+ if isinstance(obj, bytes):
2313
+ return base64.b64encode(obj).decode("utf-8")
2314
+ return super().default(obj)
2315
+
2316
+
2317
+ def _dataclass_to_dict(obj):
2318
+ if isinstance(obj, _Union):
2319
+ return {obj.type: _dataclass_to_dict(obj.value)}
2320
+ elif dataclasses.is_dataclass(obj):
2321
+ return {
2322
+ f.name: _dataclass_to_dict(getattr(obj, f.name))
2323
+ for f in dataclasses.fields(obj)
2324
+ if not (f.default is None and getattr(obj, f.name) is None)
2325
+ }
2326
+ elif isinstance(obj, list):
2327
+ return [_dataclass_to_dict(x) for x in obj]
2328
+ elif isinstance(obj, tuple):
2329
+ return tuple(_dataclass_to_dict(x) for x in obj)
2330
+ elif isinstance(obj, dict):
2331
+ return {k: _dataclass_to_dict(v) for k, v in obj.items()}
2332
+ else:
2333
+ return obj
2334
+
2335
+
2336
+ def serialize(
2337
+ exported_program: ep.ExportedProgram,
2338
+ opset_version: Optional[Dict[str, int]] = None,
2339
+ ) -> SerializedArtifact:
2340
+ serialized_program = ExportedProgramSerializer(opset_version).serialize(
2341
+ exported_program
2342
+ )
2343
+ assert isinstance(serialized_program.exported_program, ExportedProgram)
2344
+
2345
+ json_program = json.dumps(
2346
+ _dataclass_to_dict(serialized_program.exported_program), cls=EnumEncoder
2347
+ )
2348
+ json_bytes = json_program.encode("utf-8")
2349
+ artifact = SerializedArtifact(
2350
+ json_bytes,
2351
+ serialized_program.state_dict,
2352
+ serialized_program.constants,
2353
+ serialized_program.example_inputs,
2354
+ )
2355
+ return artifact
2356
+
2357
+
2358
+ def _dict_to_dataclass(cls, data):
2359
+ assert not isinstance(cls, str), f"Unresolved class type: '{cls}'."
2360
+ if typing.get_origin(cls) == typing.Union and type(None) in typing.get_args(cls):
2361
+ if data is None:
2362
+ return None
2363
+ ty_args = typing.get_args(cls)
2364
+ assert len(ty_args) == 2
2365
+ return _dict_to_dataclass(ty_args[0], data)
2366
+ elif isinstance(cls, type) and issubclass(cls, _Union):
2367
+ assert isinstance(data, dict)
2368
+ assert len(data) == 1
2369
+ _type = next(iter(data.keys()))
2370
+ _value = next(iter(data.values()))
2371
+ assert isinstance(_type, str)
2372
+ field_type = cls.__annotations__[_type]
2373
+ return cls.create(**{_type: _dict_to_dataclass(field_type, _value)})
2374
+ elif dataclasses.is_dataclass(cls):
2375
+ obj = cls(**data) # type: ignore[assignment]
2376
+ type_hints = typing.get_type_hints(cls)
2377
+ for f in dataclasses.fields(cls):
2378
+ name = f.name
2379
+ new_field_obj = _dict_to_dataclass(type_hints[name], getattr(obj, name))
2380
+ setattr(obj, name, new_field_obj)
2381
+ return obj
2382
+ elif isinstance(data, list):
2383
+ if len(data) == 0:
2384
+ return data
2385
+ d_type = typing.get_args(cls)[0]
2386
+ return [_dict_to_dataclass(d_type, d) for d in data]
2387
+ elif isinstance(data, dict):
2388
+ v_type = typing.get_args(cls)[1]
2389
+ return {k: _dict_to_dataclass(v_type, v) for k, v in data.items()}
2390
+ return data
2391
+
2392
+
2393
+ def deserialize(
2394
+ artifact: SerializedArtifact,
2395
+ ) -> ep.ExportedProgram:
2396
+ assert isinstance(artifact.exported_program, bytes)
2397
+ exported_program_str = artifact.exported_program.decode("utf-8")
2398
+ exported_program_dict = json.loads(exported_program_str)
2399
+ serialized_exported_program = _dict_to_dataclass(
2400
+ ExportedProgram, exported_program_dict
2401
+ )
2402
+ return ExportedProgramDeserializer().deserialize(
2403
+ serialized_exported_program,
2404
+ artifact.state_dict,
2405
+ artifact.constants,
2406
+ artifact.example_inputs,
2407
+ )
2408
+
2409
+
2410
+ def _canonicalize_graph(
2411
+ sorted_inputs, sorted_outputs, graph
2412
+ ) -> Tuple[Graph, Dict[str, str]]:
2413
+ def _get_argument(a: Argument):
2414
+ if a.type == "as_none":
2415
+ return None
2416
+ elif a.type == "as_tensor":
2417
+ return a.as_tensor
2418
+ elif a.type == "as_tensors":
2419
+ return a.as_tensors
2420
+ elif a.type == "as_int":
2421
+ return None
2422
+ elif a.type == "as_ints":
2423
+ return None
2424
+ elif a.type == "as_float":
2425
+ return None
2426
+ elif a.type == "as_floats":
2427
+ return None
2428
+ elif a.type == "as_string":
2429
+ return None
2430
+ elif a.type == "as_strings":
2431
+ return None
2432
+ elif a.type == "as_sym_int":
2433
+ return a.as_sym_int
2434
+ elif a.type == "as_sym_ints":
2435
+ return a.as_sym_ints
2436
+ elif a.type == "as_scalar_type":
2437
+ return None
2438
+ elif a.type == "as_memory_format":
2439
+ return None
2440
+ elif a.type == "as_layout":
2441
+ return None
2442
+ elif a.type == "as_device":
2443
+ return None
2444
+ elif a.type == "as_bool":
2445
+ return None
2446
+ elif a.type == "as_bools":
2447
+ return None
2448
+ elif a.type == "as_sym_bool":
2449
+ return a.as_sym_bool
2450
+ elif a.type == "as_sym_bools":
2451
+ return a.as_sym_bools
2452
+ elif a.type == "as_graph":
2453
+ return None
2454
+ elif a.type == "as_optional_tensors":
2455
+ return a.as_optional_tensors
2456
+ elif a.type == "as_custom_obj":
2457
+ return None
2458
+ elif a.type == "as_operator":
2459
+ return None
2460
+ else:
2461
+ raise AssertionError(f"Unknown input type to the ExportedProgram: {a}")
2462
+
2463
+ # Stage 1: Reorder named items.
2464
+ def for_args(f, a):
2465
+ assert isinstance(a, Argument)
2466
+ pytree.tree_map(f, _get_argument(a))
2467
+
2468
+ def sort_nodes(nodes):
2469
+ @dataclass
2470
+ class Edges:
2471
+ outs: List[int]
2472
+ ins: int
2473
+
2474
+ graph_inputs: Set[str] = set()
2475
+ def_table: Dict[str, int] = {}
2476
+ edges: Dict[int, Edges] = {}
2477
+ candidates: List[Tuple[str, List[Tuple[str, List[int]]], int]] = []
2478
+ rank: Dict[str, int] = {}
2479
+ ret: List[Node] = []
2480
+
2481
+ def get_name(a) -> Optional[str]:
2482
+ if a is None:
2483
+ return None
2484
+ if isinstance(a, TensorArgument):
2485
+ return a.name
2486
+ elif isinstance(a, (SymIntArgument, SymBoolArgument)):
2487
+ if a.type == "as_name":
2488
+ return a.as_name
2489
+ elif a.type in ("as_int", "as_bool"):
2490
+ return None
2491
+ else:
2492
+ raise AssertionError(f"Unknown argument type: {a}")
2493
+ elif isinstance(a, OptionalTensorArgument):
2494
+ if a.type == "as_tensor":
2495
+ return a.as_tensor.name
2496
+ elif a.type == "as_none":
2497
+ return None
2498
+ else:
2499
+ raise AssertionError(f"Unknown optional tensor type: {a}")
2500
+ else:
2501
+ raise AssertionError(f"Unknown argument type: {a}")
2502
+
2503
+ for i in sorted_inputs:
2504
+
2505
+ def add_input(a):
2506
+ if s := get_name(a):
2507
+ graph_inputs.add(s)
2508
+
2509
+ for_args(add_input, i)
2510
+
2511
+ for idx, node in enumerate(nodes):
2512
+
2513
+ def add_def(a):
2514
+ if s := get_name(a):
2515
+ assert s not in def_table
2516
+ def_table[s] = idx
2517
+
2518
+ for o in node.outputs:
2519
+ for_args(add_def, o)
2520
+
2521
+ edges[idx] = Edges([], 0)
2522
+
2523
+ for idx, user in enumerate(nodes):
2524
+
2525
+ def add_edge(a):
2526
+ if s := get_name(a):
2527
+ if s not in def_table:
2528
+ assert s in graph_inputs
2529
+ return
2530
+ src = def_table[s]
2531
+ edges[src].outs.append(idx)
2532
+ edges[idx].ins += 1
2533
+
2534
+ for i in user.inputs:
2535
+ for_args(add_edge, i.arg)
2536
+
2537
+ def add_rank(a):
2538
+ if s := get_name(a):
2539
+ assert s not in rank
2540
+ rank[s] = len(rank)
2541
+
2542
+ def get_rank(a):
2543
+ if s := get_name(a):
2544
+ return rank[s]
2545
+ else:
2546
+ return -1
2547
+
2548
+ for i in sorted_inputs:
2549
+ for_args(add_rank, i)
2550
+
2551
+ def add_candidate(idx: int):
2552
+ def get_ranks(i):
2553
+ ranks = []
2554
+ for_args(lambda x: ranks.append(get_rank(x)), i)
2555
+ return ranks
2556
+
2557
+ node = nodes[idx]
2558
+ args_rank = [(a.name, get_ranks(a.arg)) for a in node.inputs]
2559
+ heapq.heappush(candidates, (node.target, args_rank, idx))
2560
+
2561
+ for idx, e in edges.items():
2562
+ if e.ins == 0:
2563
+ add_candidate(idx)
2564
+
2565
+ while len(candidates) > 0:
2566
+ _, _, idx = heapq.heappop(candidates)
2567
+ node = nodes[idx]
2568
+ for o in node.outputs:
2569
+ for_args(add_rank, o)
2570
+ ret.append(node)
2571
+ assert idx in edges
2572
+ for user in edges[idx].outs:
2573
+ e = edges[user]
2574
+ assert e.ins > 0
2575
+ e.ins -= 1
2576
+ if e.ins == 0:
2577
+ add_candidate(user)
2578
+ edges[idx].outs.clear()
2579
+
2580
+ return ret
2581
+
2582
+ sorted_nodes = sort_nodes(graph.nodes)
2583
+ assert len(sorted_nodes) == len(graph.nodes)
2584
+
2585
+ # Stage 2: Rename nodes.
2586
+ name_table: Dict[str, str] = {}
2587
+
2588
+ def rename_def(a):
2589
+ def _rename(arg_name, values):
2590
+ new_name = f"_{len(name_table)}"
2591
+ assert arg_name not in name_table
2592
+ name_table[arg_name] = new_name
2593
+ assert arg_name in values
2594
+ values[new_name] = values.pop(arg_name)
2595
+ return new_name
2596
+
2597
+ if a is None:
2598
+ return
2599
+ if isinstance(a, TensorArgument):
2600
+ a.name = _rename(a.name, graph.tensor_values)
2601
+ elif isinstance(a, SymIntArgument):
2602
+ if a.type == "as_name":
2603
+ a.as_name = _rename(a.as_name, graph.sym_int_values)
2604
+ elif isinstance(a, SymBoolArgument):
2605
+ if a.type == "as_name":
2606
+ a.as_name = _rename(a.as_name, graph.sym_bool_values)
2607
+ else:
2608
+ raise AssertionError(f"Unknown argument type: {a}")
2609
+
2610
+ def replace_use(a):
2611
+ if a is None:
2612
+ return
2613
+ if isinstance(a, TensorArgument):
2614
+ a.name = name_table.get(a.name, a.name)
2615
+ elif isinstance(a, SymIntArgument):
2616
+ if a.type == "as_name":
2617
+ a.as_name = name_table.get(a.as_name, a.as_name)
2618
+ elif isinstance(a, SymBoolArgument):
2619
+ if a.type == "as_name":
2620
+ a.as_name = name_table.get(a.as_name, a.as_name)
2621
+ elif isinstance(a, OptionalTensorArgument):
2622
+ if a.type == "as_tensor":
2623
+ a.as_tensor.name = name_table.get(a.as_tensor.name, a.as_tensor.name)
2624
+ else:
2625
+ raise AssertionError(f"Unknown argument type: {a}")
2626
+
2627
+ for i in sorted_inputs:
2628
+ for_args(rename_def, i)
2629
+
2630
+ for n in sorted_nodes:
2631
+ for o in n.outputs:
2632
+ for_args(rename_def, o)
2633
+
2634
+ for n in sorted_nodes:
2635
+ for i in n.inputs:
2636
+ for_args(replace_use, i.arg)
2637
+
2638
+ for o in sorted_outputs:
2639
+ for_args(replace_use, o)
2640
+
2641
+ # Stage 3: Remove unstable fields.
2642
+ for n in sorted_nodes:
2643
+ n.metadata.clear()
2644
+
2645
+ # Stage 4: Aggregate values.
2646
+ sorted_tensor_values = dict(
2647
+ sorted(graph.tensor_values.items(), key=operator.itemgetter(0))
2648
+ )
2649
+ sorted_sym_int_values = dict(
2650
+ sorted(graph.sym_int_values.items(), key=operator.itemgetter(0))
2651
+ )
2652
+ sorted_sym_bool_values = dict(
2653
+ sorted(graph.sym_bool_values.items(), key=operator.itemgetter(0))
2654
+ )
2655
+
2656
+ # Stage 5: Recurse in subgraphs.
2657
+ counter = 0
2658
+ for node in sorted_nodes:
2659
+ for i in node.inputs:
2660
+ a = i.arg
2661
+ if a.type == "as_graph":
2662
+ a.as_graph.graph = _canonicalize_graph(
2663
+ a.as_graph.graph.inputs, a.as_graph.graph.outputs, a.as_graph.graph
2664
+ )
2665
+ a.as_graph.name = f"_g{counter}"
2666
+ counter += 1
2667
+
2668
+ graph = Graph(
2669
+ inputs=sorted_inputs,
2670
+ outputs=sorted_outputs,
2671
+ nodes=sorted_nodes,
2672
+ tensor_values=sorted_tensor_values,
2673
+ sym_int_values=sorted_sym_int_values,
2674
+ sym_bool_values=sorted_sym_bool_values,
2675
+ is_single_tensor_return=graph.is_single_tensor_return,
2676
+ )
2677
+ return graph, name_table
2678
+
2679
+
2680
+ def canonicalize(ep: ExportedProgram) -> ExportedProgram:
2681
+ """
2682
+ Normalize a serialized ExportedProgram, so that different eager program which
2683
+ shares the same semantics can get a single representation on disk.
2684
+
2685
+ This function canonicalizes an ExportedProgram by:
2686
+
2687
+ 1. Sorting nodes in topological order.
2688
+ 2. Rename nodes to have unique names.
2689
+ 3. Remove unstable fields.
2690
+ 4. Aggregate the above program fields.
2691
+ 5. Recurse in subgraphs.
2692
+
2693
+ Args:
2694
+ ep (ExportedProgram): The ExportedProgram to canonicalize.
2695
+
2696
+ Returns:
2697
+ ExportedProgram: The canonicalized exported program.
2698
+ """
2699
+ ep = copy.deepcopy(ep)
2700
+
2701
+ opset_version = dict(sorted(ep.opset_version.items(), key=operator.itemgetter(0)))
2702
+ range_constraints = dict(
2703
+ sorted(ep.range_constraints.items(), key=operator.itemgetter(0))
2704
+ )
2705
+ module_call_graph = sorted(ep.graph_module.module_call_graph, key=lambda x: x.fqn)
2706
+ signature = ep.graph_module.signature
2707
+ graph = ep.graph_module.graph
2708
+
2709
+ assert len(graph.inputs) == len(signature.input_specs)
2710
+ assert len(graph.outputs) == len(signature.output_specs)
2711
+
2712
+ def rank_input(inp) -> Tuple[int, Optional[str], int]:
2713
+ idx, (arg, spec) = inp
2714
+ assert isinstance(spec, InputSpec)
2715
+ if spec.type == "user_input":
2716
+ return 5, None, idx
2717
+ elif spec.type == "parameter":
2718
+ return 1, spec.parameter.parameter_name, idx
2719
+ elif spec.type == "buffer":
2720
+ return 2, spec.buffer.buffer_name, idx
2721
+ elif spec.type == "tensor_constant":
2722
+ return 3, spec.tensor_constant.tensor_constant_name, idx
2723
+ elif spec.type == "custom_obj":
2724
+ return 4, spec.custom_obj.custom_obj_name, idx
2725
+ elif spec.type == "token":
2726
+ return 0, None, idx
2727
+ elif spec.type == "constant_input":
2728
+ return 6, spec.constant_input.name, idx
2729
+ else:
2730
+ raise AssertionError(f"Unknown input type: {spec}")
2731
+
2732
+ def rank_output(out) -> Tuple[int, Optional[str], int]:
2733
+ idx, (arg, spec) = out
2734
+ assert isinstance(spec, OutputSpec)
2735
+ if spec.type == "user_output":
2736
+ return 3, None, idx
2737
+ elif spec.type == "loss_output":
2738
+ return 3, None, idx
2739
+ elif spec.type == "buffer_mutation":
2740
+ return 1, spec.buffer_mutation.buffer_name, idx
2741
+ elif spec.type == "gradient_to_parameter":
2742
+ return 4, spec.gradient_to_parameter.parameter_name, idx
2743
+ elif spec.type == "gradient_to_user_input":
2744
+ return 5, None, idx
2745
+ elif spec.type == "user_input_mutation":
2746
+ return 2, None, idx
2747
+ elif spec.type == "token":
2748
+ return 0, None, idx
2749
+ else:
2750
+ raise AssertionError(f"Unknown output type: {spec}")
2751
+
2752
+ sorted_ins = sorted(
2753
+ enumerate(zip(graph.inputs, signature.input_specs)), key=rank_input
2754
+ )
2755
+ sorted_inputs, input_specs = zip(*(i for idx, i in sorted_ins)) # type: ignore[assignment]
2756
+
2757
+ sorted_outs = sorted(
2758
+ enumerate(zip(graph.outputs, signature.output_specs)), key=rank_output
2759
+ )
2760
+ sorted_outputs, output_specs = zip(*(i for idx, i in sorted_outs)) # type: ignore[assignment]
2761
+
2762
+ sorted_graph, replace_table = _canonicalize_graph(
2763
+ sorted_inputs, sorted_outputs, graph
2764
+ )
2765
+
2766
+ def replace_input(inp):
2767
+ assert isinstance(spec, InputSpec)
2768
+ if spec.type == "user_input":
2769
+ arg = spec.user_input.arg
2770
+ if arg.type == "as_tensor":
2771
+ t = arg.as_tensor
2772
+ t.name = replace_table[t.name]
2773
+ elif arg.type == "as_sym_int":
2774
+ s = arg.as_sym_int
2775
+ if s.type == "as_name":
2776
+ s.as_name = replace_table[s.as_name]
2777
+ elif s.type == "as_int":
2778
+ pass
2779
+ else:
2780
+ raise AssertionError(f"Unknown sym_int type: {s}")
2781
+ elif arg.type in (
2782
+ "as_none",
2783
+ "as_bool",
2784
+ "as_int",
2785
+ "as_float",
2786
+ "as_string",
2787
+ "as_custom_obj",
2788
+ ):
2789
+ return
2790
+ else:
2791
+ raise AssertionError(f"Unknown input type: {arg}")
2792
+ elif spec.type == "parameter":
2793
+ t = spec.parameter.arg
2794
+ t.name = replace_table[t.name]
2795
+ elif spec.type == "buffer":
2796
+ t = spec.buffer.arg
2797
+ t.name = replace_table[t.name]
2798
+ elif spec.type == "tensor_constant":
2799
+ t = spec.tensor_constant.arg
2800
+ t.name = replace_table[t.name]
2801
+ elif spec.type == "custom_obj":
2802
+ return
2803
+ elif spec.type == "token":
2804
+ tok = spec.token.arg
2805
+ tok.name = replace_table[tok.name]
2806
+ elif spec.type == "constant_input":
2807
+ return
2808
+ else:
2809
+ raise AssertionError(f"Unknown input type: {spec}")
2810
+
2811
+ def replace_output(out):
2812
+ assert isinstance(spec, OutputSpec)
2813
+ if spec.type == "user_output":
2814
+ arg = spec.user_output.arg
2815
+ if arg.type == "as_tensor":
2816
+ t = arg.as_tensor
2817
+ t.name = replace_table[t.name]
2818
+ elif arg.type == "as_sym_int":
2819
+ s = arg.as_sym_int
2820
+ if s.type == "as_name":
2821
+ s.as_name = replace_table[s.as_name]
2822
+ elif s.type == "as_int":
2823
+ pass
2824
+ else:
2825
+ raise AssertionError(f"Unknown sym_int type: {s}")
2826
+ elif arg.type in ("as_none", "as_int", "as_float", "as_string"):
2827
+ return
2828
+ else:
2829
+ raise AssertionError(f"Unknown input type: {arg}")
2830
+ elif spec.type == "loss_output":
2831
+ t = spec.loss_output.arg
2832
+ t.name = replace_table[t.name]
2833
+ elif spec.type == "buffer_mutation":
2834
+ t = spec.buffer_mutation.arg
2835
+ t.name = replace_table[t.name]
2836
+ elif spec.type == "gradient_to_parameter":
2837
+ t = spec.gradient_to_parameter.arg
2838
+ t.name = replace_table[t.name]
2839
+ elif spec.type == "gradient_to_user_input":
2840
+ g = spec.gradient_to_user_input
2841
+ g.arg.name = replace_table[g.arg.name]
2842
+ g.user_input_name = replace_table[g.user_input_name]
2843
+ elif spec.type == "user_input_mutation":
2844
+ u = spec.user_input_mutation
2845
+ u.arg.name = replace_table[u.arg.name]
2846
+ u.user_input_name = replace_table[u.user_input_name]
2847
+ elif spec.type == "token":
2848
+ tok = spec.token.arg
2849
+ tok.name = replace_table[tok.name]
2850
+ else:
2851
+ raise AssertionError(f"Unknown output type: {spec}")
2852
+
2853
+ for spec in input_specs:
2854
+ replace_input(spec)
2855
+
2856
+ for spec in output_specs:
2857
+ replace_output(spec)
2858
+
2859
+ return ExportedProgram(
2860
+ graph_module=GraphModule(
2861
+ graph=sorted_graph,
2862
+ signature=GraphSignature(
2863
+ input_specs=list(input_specs),
2864
+ output_specs=list(output_specs),
2865
+ ),
2866
+ module_call_graph=module_call_graph,
2867
+ ),
2868
+ opset_version=opset_version,
2869
+ range_constraints=range_constraints,
2870
+ schema_version=ep.schema_version,
2871
+ dialect=ep.dialect,
2872
+ )