mindstudio-probe 1.0.3__py3-none-any.whl → 1.1.0__py3-none-any.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 (278) hide show
  1. {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/LICENSE +201 -201
  2. {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/METADATA +36 -34
  3. mindstudio_probe-1.1.0.dist-info/RECORD +287 -0
  4. {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/WHEEL +1 -1
  5. {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/entry_points.txt +1 -0
  6. msprobe/README.md +131 -237
  7. msprobe/__init__.py +16 -1
  8. msprobe/{config/config.json → config.json} +47 -49
  9. msprobe/core/advisor/advisor.py +124 -124
  10. msprobe/core/advisor/advisor_const.py +58 -59
  11. msprobe/core/advisor/advisor_result.py +58 -58
  12. msprobe/core/common/const.py +402 -318
  13. msprobe/core/common/exceptions.py +99 -99
  14. msprobe/core/common/{file_check.py → file_utils.py} +523 -283
  15. msprobe/core/common/inplace_op_checker.py +38 -0
  16. msprobe/core/common/inplace_ops.yaml +251 -0
  17. msprobe/core/common/log.py +86 -69
  18. msprobe/core/common/utils.py +371 -616
  19. msprobe/core/common_config.py +78 -71
  20. msprobe/core/compare/acc_compare.py +472 -298
  21. msprobe/core/compare/check.py +180 -95
  22. msprobe/core/compare/compare_cli.py +69 -49
  23. msprobe/core/compare/highlight.py +259 -222
  24. msprobe/core/compare/multiprocessing_compute.py +174 -149
  25. msprobe/core/compare/npy_compare.py +310 -295
  26. msprobe/core/compare/utils.py +464 -429
  27. msprobe/core/data_dump/data_collector.py +153 -144
  28. msprobe/core/data_dump/data_processor/base.py +337 -293
  29. msprobe/core/data_dump/data_processor/factory.py +76 -59
  30. msprobe/core/data_dump/data_processor/mindspore_processor.py +192 -198
  31. msprobe/core/data_dump/data_processor/pytorch_processor.py +383 -389
  32. msprobe/core/data_dump/json_writer.py +117 -116
  33. msprobe/core/data_dump/scope.py +194 -178
  34. msprobe/core/grad_probe/constant.py +74 -70
  35. msprobe/core/grad_probe/grad_compare.py +170 -175
  36. msprobe/core/grad_probe/utils.py +77 -52
  37. msprobe/docs/01.installation.md +99 -0
  38. msprobe/docs/02.config_introduction.md +137 -0
  39. msprobe/docs/03.config_examples.md +237 -0
  40. msprobe/docs/04.acl_config_examples.md +78 -0
  41. msprobe/docs/05.data_dump_PyTorch.md +326 -0
  42. msprobe/docs/06.data_dump_MindSpore.md +285 -0
  43. msprobe/docs/07.accuracy_checker_PyTorch.md +297 -0
  44. msprobe/docs/08.accuracy_checker_online_PyTorch.md +238 -0
  45. msprobe/docs/09.accuracy_checker_MindSpore.md +68 -0
  46. msprobe/docs/10.accuracy_compare_PyTorch.md +327 -0
  47. msprobe/docs/11.accuracy_compare_MindSpore.md +333 -0
  48. msprobe/docs/12.overflow_check_PyTorch.md +79 -0
  49. msprobe/docs/13.overflow_check_MindSpore.md +31 -0
  50. msprobe/{pytorch/doc/parse_tool.md → docs/14.data_parse_PyTorch.md} +283 -286
  51. msprobe/docs/15.free_benchmarking_PyTorch.md +170 -0
  52. msprobe/docs/16.free_benchmarking_MindSpore.md +140 -0
  53. msprobe/{doc/grad_probe/grad_probe.md → docs/17.grad_probe.md} +205 -207
  54. msprobe/{pytorch/doc//321/205/320/254/320/270/321/207/342/225/221/342/224/220/321/207/342/226/223/342/225/233/321/205/342/225/221/320/266/321/206/320/277/320/244/321/205/320/277/342/225/243.md → docs/18.online_dispatch.md} +89 -90
  55. msprobe/docs/FAQ.md +189 -0
  56. msprobe/docs/S02.report_free_benchmarking_validation_performance_baseline.md +146 -0
  57. msprobe/docs/img/free_benchmark_framework.png +0 -0
  58. msprobe/docs/img/ms_dump.png +0 -0
  59. msprobe/docs/img/ms_layer.png +0 -0
  60. msprobe/docs/img/pt_dump.png +0 -0
  61. msprobe/mindspore/__init__.py +2 -1
  62. msprobe/mindspore/api_accuracy_checker/api_accuracy_checker.py +278 -245
  63. msprobe/mindspore/api_accuracy_checker/api_info.py +76 -69
  64. msprobe/mindspore/api_accuracy_checker/api_runner.py +155 -151
  65. msprobe/mindspore/api_accuracy_checker/base_compare_algorithm.py +196 -196
  66. msprobe/mindspore/api_accuracy_checker/cmd_parser.py +6 -0
  67. msprobe/mindspore/api_accuracy_checker/compute_element.py +238 -223
  68. msprobe/mindspore/api_accuracy_checker/main.py +8 -15
  69. msprobe/mindspore/api_accuracy_checker/type_mapping.py +113 -113
  70. msprobe/mindspore/api_accuracy_checker/utils.py +79 -62
  71. msprobe/mindspore/cell_processor.py +58 -34
  72. msprobe/mindspore/common/const.py +108 -87
  73. msprobe/mindspore/common/log.py +37 -37
  74. msprobe/mindspore/common/utils.py +97 -57
  75. msprobe/mindspore/compare/distributed_compare.py +62 -75
  76. msprobe/mindspore/compare/layer_mapping.py +146 -0
  77. msprobe/mindspore/compare/modify_mapping.py +107 -0
  78. msprobe/mindspore/compare/ms_compare.py +357 -117
  79. msprobe/mindspore/compare/ms_graph_compare.py +364 -317
  80. msprobe/mindspore/compare/ms_to_pt_api.yaml +399 -399
  81. msprobe/mindspore/debugger/debugger_config.py +69 -74
  82. msprobe/mindspore/debugger/precision_debugger.py +150 -107
  83. msprobe/mindspore/dump/dump_tool_factory.py +50 -35
  84. msprobe/mindspore/dump/hook_cell/api_registry.py +128 -104
  85. msprobe/mindspore/dump/hook_cell/hook_cell.py +55 -53
  86. msprobe/mindspore/dump/hook_cell/primitive_hooks.py +206 -0
  87. msprobe/mindspore/dump/hook_cell/support_wrap_ops.yaml +994 -925
  88. msprobe/mindspore/dump/hook_cell/wrap_api.py +121 -0
  89. msprobe/mindspore/dump/jit_dump.py +96 -56
  90. msprobe/mindspore/dump/kernel_graph_dump.py +75 -60
  91. msprobe/mindspore/dump/kernel_kbyk_dump.py +79 -65
  92. msprobe/mindspore/free_benchmark/api_pynative_self_check.py +131 -116
  93. msprobe/mindspore/free_benchmark/common/config.py +27 -12
  94. msprobe/mindspore/free_benchmark/common/handler_params.py +32 -17
  95. msprobe/mindspore/free_benchmark/common/utils.py +85 -71
  96. msprobe/mindspore/free_benchmark/data/support_wrap_ops.yaml +842 -842
  97. msprobe/mindspore/free_benchmark/decorator/dec_forward.py +57 -42
  98. msprobe/mindspore/free_benchmark/decorator/decorator_factory.py +122 -107
  99. msprobe/mindspore/free_benchmark/handler/base_handler.py +105 -90
  100. msprobe/mindspore/free_benchmark/handler/check_handler.py +56 -41
  101. msprobe/mindspore/free_benchmark/handler/fix_handler.py +51 -36
  102. msprobe/mindspore/free_benchmark/handler/handler_factory.py +36 -21
  103. msprobe/mindspore/free_benchmark/perturbation/add_noise.py +82 -67
  104. msprobe/mindspore/free_benchmark/perturbation/base_perturbation.py +36 -21
  105. msprobe/mindspore/free_benchmark/perturbation/bit_noise.py +78 -63
  106. msprobe/mindspore/free_benchmark/perturbation/exchange_value.py +77 -0
  107. msprobe/mindspore/free_benchmark/perturbation/improve_precision.py +49 -34
  108. msprobe/mindspore/free_benchmark/perturbation/no_change.py +27 -12
  109. msprobe/mindspore/free_benchmark/perturbation/perturbation_factory.py +44 -27
  110. msprobe/mindspore/free_benchmark/self_check_tool_factory.py +48 -33
  111. msprobe/mindspore/grad_probe/global_context.py +100 -91
  112. msprobe/mindspore/grad_probe/grad_analyzer.py +231 -231
  113. msprobe/mindspore/grad_probe/grad_monitor.py +27 -27
  114. msprobe/mindspore/grad_probe/grad_stat_csv.py +131 -131
  115. msprobe/mindspore/grad_probe/hook.py +94 -92
  116. msprobe/mindspore/grad_probe/utils.py +29 -28
  117. msprobe/mindspore/ms_config.py +128 -126
  118. msprobe/mindspore/overflow_check/kernel_graph_overflow_check.py +60 -45
  119. msprobe/mindspore/overflow_check/overflow_check_tool_factory.py +49 -34
  120. msprobe/mindspore/runtime.py +4 -4
  121. msprobe/mindspore/service.py +297 -354
  122. msprobe/mindspore/task_handler_factory.py +24 -24
  123. msprobe/msprobe.py +105 -107
  124. msprobe/pytorch/__init__.py +23 -4
  125. msprobe/pytorch/api_accuracy_checker/common/config.py +70 -55
  126. msprobe/pytorch/api_accuracy_checker/common/utils.py +246 -165
  127. msprobe/pytorch/api_accuracy_checker/compare/algorithm.py +230 -213
  128. msprobe/pytorch/api_accuracy_checker/compare/api_precision_compare.py +632 -581
  129. msprobe/pytorch/api_accuracy_checker/compare/api_precision_standard.yaml +132 -132
  130. msprobe/pytorch/api_accuracy_checker/compare/api_precision_threshold.yaml +390 -390
  131. msprobe/pytorch/api_accuracy_checker/compare/compare.py +416 -381
  132. msprobe/pytorch/api_accuracy_checker/compare/compare_column.py +90 -73
  133. msprobe/pytorch/api_accuracy_checker/compare/compare_utils.py +265 -244
  134. msprobe/pytorch/api_accuracy_checker/config.yaml +10 -10
  135. msprobe/pytorch/api_accuracy_checker/run_ut/data_generate.py +370 -332
  136. msprobe/pytorch/api_accuracy_checker/run_ut/multi_run_ut.py +221 -199
  137. msprobe/pytorch/api_accuracy_checker/run_ut/run_overflow_check.py +150 -134
  138. msprobe/pytorch/api_accuracy_checker/run_ut/run_ut.py +518 -581
  139. msprobe/pytorch/api_accuracy_checker/run_ut/run_ut_utils.py +213 -74
  140. msprobe/pytorch/api_accuracy_checker/run_ut/torch_ut_setting.json +7 -4
  141. msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/attl.py +218 -202
  142. msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/client.py +370 -324
  143. msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/device_dispatch.py +227 -204
  144. msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/dump_dispatch.py +110 -0
  145. msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/server.py +244 -218
  146. msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/torch_ops_config.yaml +63 -0
  147. msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/utils.py +44 -0
  148. msprobe/pytorch/bench_functions/__init__.py +30 -15
  149. msprobe/pytorch/bench_functions/apply_adam_w.py +43 -28
  150. msprobe/pytorch/bench_functions/confusion_transpose.py +34 -19
  151. msprobe/pytorch/bench_functions/fast_gelu.py +70 -55
  152. msprobe/pytorch/bench_functions/layer_norm_eval.py +21 -6
  153. msprobe/pytorch/bench_functions/linear.py +27 -12
  154. msprobe/pytorch/bench_functions/matmul_backward.py +63 -48
  155. msprobe/pytorch/bench_functions/npu_fusion_attention.py +538 -421
  156. msprobe/pytorch/bench_functions/rms_norm.py +30 -15
  157. msprobe/pytorch/bench_functions/rotary_mul.py +71 -52
  158. msprobe/pytorch/bench_functions/scaled_mask_softmax.py +41 -26
  159. msprobe/pytorch/bench_functions/swiglu.py +70 -55
  160. msprobe/pytorch/common/__init__.py +17 -2
  161. msprobe/pytorch/common/compare_script.template +14 -14
  162. msprobe/pytorch/common/log.py +33 -32
  163. msprobe/pytorch/common/parse_json.py +54 -39
  164. msprobe/pytorch/common/utils.py +310 -300
  165. msprobe/pytorch/compare/distributed_compare.py +66 -66
  166. msprobe/pytorch/compare/mapping.yaml +607 -607
  167. msprobe/pytorch/compare/match.py +49 -33
  168. msprobe/pytorch/compare/pt_compare.py +82 -40
  169. msprobe/pytorch/debugger/debugger_config.py +108 -95
  170. msprobe/pytorch/debugger/precision_debugger.py +173 -125
  171. msprobe/pytorch/free_benchmark/__init__.py +23 -8
  172. msprobe/pytorch/free_benchmark/common/constant.py +70 -70
  173. msprobe/pytorch/free_benchmark/common/counter.py +71 -71
  174. msprobe/pytorch/free_benchmark/common/enums.py +65 -37
  175. msprobe/pytorch/free_benchmark/common/params.py +144 -129
  176. msprobe/pytorch/free_benchmark/common/utils.py +118 -102
  177. msprobe/pytorch/free_benchmark/compare/grad_saver.py +200 -179
  178. msprobe/pytorch/free_benchmark/compare/single_benchmark.py +119 -104
  179. msprobe/pytorch/free_benchmark/main.py +120 -105
  180. msprobe/pytorch/free_benchmark/perturbed_layers/base_layer.py +28 -13
  181. msprobe/pytorch/free_benchmark/perturbed_layers/layer_factory.py +56 -41
  182. msprobe/pytorch/free_benchmark/perturbed_layers/npu/add_noise.py +105 -90
  183. msprobe/pytorch/free_benchmark/perturbed_layers/npu/bit_noise.py +119 -104
  184. msprobe/pytorch/free_benchmark/perturbed_layers/npu/change_value.py +87 -63
  185. msprobe/pytorch/free_benchmark/perturbed_layers/npu/improve_precision.py +83 -68
  186. msprobe/pytorch/free_benchmark/perturbed_layers/npu/no_change.py +43 -28
  187. msprobe/pytorch/free_benchmark/perturbed_layers/npu/npu_base_layser.py +60 -45
  188. msprobe/pytorch/free_benchmark/perturbed_layers/run_cpu.py +34 -19
  189. msprobe/pytorch/free_benchmark/result_handlers/base_handler.py +256 -217
  190. msprobe/pytorch/free_benchmark/result_handlers/check_handler.py +54 -39
  191. msprobe/pytorch/free_benchmark/result_handlers/fix_handler.py +38 -23
  192. msprobe/pytorch/free_benchmark/result_handlers/handler_factory.py +45 -30
  193. msprobe/pytorch/free_benchmark/result_handlers/preheat_handler.py +185 -170
  194. msprobe/pytorch/function_factory.py +91 -75
  195. msprobe/pytorch/functional/module_dump.py +84 -0
  196. msprobe/pytorch/grad_probe/grad_monitor.py +91 -90
  197. msprobe/pytorch/grad_probe/grad_stat_csv.py +128 -128
  198. msprobe/pytorch/hook_module/__init__.py +16 -1
  199. msprobe/pytorch/hook_module/api_registry.py +166 -161
  200. msprobe/pytorch/hook_module/hook_module.py +118 -120
  201. msprobe/pytorch/hook_module/support_wrap_ops.yaml +1879 -1877
  202. msprobe/pytorch/hook_module/utils.py +28 -29
  203. msprobe/pytorch/hook_module/wrap_aten.py +111 -110
  204. msprobe/pytorch/hook_module/wrap_distributed.py +77 -78
  205. msprobe/pytorch/hook_module/wrap_functional.py +104 -105
  206. msprobe/pytorch/hook_module/wrap_npu_custom.py +85 -84
  207. msprobe/pytorch/hook_module/wrap_tensor.py +69 -71
  208. msprobe/pytorch/hook_module/wrap_torch.py +84 -86
  209. msprobe/pytorch/hook_module/wrap_vf.py +60 -62
  210. msprobe/pytorch/module_processer.py +153 -138
  211. msprobe/pytorch/online_dispatch/__init__.py +20 -20
  212. msprobe/pytorch/online_dispatch/compare.py +235 -236
  213. msprobe/pytorch/online_dispatch/dispatch.py +271 -271
  214. msprobe/pytorch/online_dispatch/dump_compare.py +155 -156
  215. msprobe/pytorch/online_dispatch/single_compare.py +391 -391
  216. msprobe/pytorch/online_dispatch/torch_ops_config.yaml +57 -49
  217. msprobe/pytorch/online_dispatch/utils.py +127 -146
  218. msprobe/pytorch/parse.py +19 -4
  219. msprobe/pytorch/parse_tool/cli.py +31 -32
  220. msprobe/pytorch/parse_tool/lib/compare.py +259 -271
  221. msprobe/pytorch/parse_tool/lib/config.py +52 -52
  222. msprobe/pytorch/parse_tool/lib/file_desc.py +31 -31
  223. msprobe/pytorch/parse_tool/lib/interactive_cli.py +102 -102
  224. msprobe/pytorch/parse_tool/lib/parse_exception.py +54 -54
  225. msprobe/pytorch/parse_tool/lib/parse_tool.py +161 -158
  226. msprobe/pytorch/parse_tool/lib/utils.py +320 -321
  227. msprobe/pytorch/parse_tool/lib/visualization.py +85 -91
  228. msprobe/pytorch/pt_config.py +317 -187
  229. msprobe/pytorch/service.py +311 -252
  230. mindstudio_probe-1.0.3.dist-info/RECORD +0 -272
  231. msprobe/config/README.md +0 -539
  232. msprobe/mindspore/doc/compare.md +0 -58
  233. msprobe/mindspore/doc/dump.md +0 -217
  234. msprobe/mindspore/dump/hook_cell/wrap_functional.py +0 -91
  235. msprobe/mindspore/dump/hook_cell/wrap_tensor.py +0 -63
  236. msprobe/pytorch/api_accuracy_checker/tensor_transport_layer/ssl_config.py +0 -10
  237. msprobe/pytorch/doc/FAQ.md +0 -193
  238. msprobe/pytorch/doc/api_accuracy_checker.md +0 -313
  239. msprobe/pytorch/doc/api_accuracy_checker_online.md +0 -187
  240. msprobe/pytorch/doc/dump.md +0 -260
  241. msprobe/pytorch/doc/msprobe/321/207/342/226/223/342/225/233/321/205/342/225/221/320/266/321/205/342/225/226/320/265/321/205/320/225/342/225/226/321/206/320/245/342/226/221/321/206/320/235/320/276dump/321/206/320/260/320/227/321/205/320/227/320/226/321/206/320/220/320/267/321/210/320/223/342/225/234/321/205/320/257/342/225/221/321/207/342/225/221/342/224/220/321/206/320/232/320/265/321/205/320/241/320/232.md +0 -182
  242. msprobe/pytorch/doc/ptdbg_ascend_compare.md +0 -240
  243. msprobe/pytorch/doc/ptdbg_ascend_overview.md +0 -68
  244. msprobe/pytorch/doc/ptdbg_ascend_quickstart.md +0 -381
  245. msprobe/pytorch/doc/run_overflow_check.md +0 -25
  246. msprobe/pytorch/doc//321/206/320/247/320/260/321/206/320/260/320/227/321/206/320/255/320/226/321/205/342/225/226/320/265/321/205/320/225/342/225/226/321/205/320/254/342/225/221/321/206/320/251/320/277/321/211/320/272/320/234/321/210/320/277/320/221/321/205/320/242/320/234/321/206/320/220/320/267/321/210/320/223/342/225/234/321/205/320/257/342/225/221/321/207/342/225/221/342/224/220/321/206/320/232/320/265/321/205/320/241/320/232.md +0 -151
  247. msprobe/pytorch/functional/data_processor.py +0 -0
  248. msprobe/pytorch/functional/dump_module.py +0 -39
  249. {mindstudio_probe-1.0.3.dist-info → mindstudio_probe-1.1.0.dist-info}/top_level.txt +0 -0
  250. /msprobe/{pytorch/doc → docs}/img/BLOOM-7B_1.png +0 -0
  251. /msprobe/{pytorch/doc → docs}/img/BLOOM-7B_2.png +0 -0
  252. /msprobe/{pytorch/doc → docs}/img/BLOOM-7B_3.png +0 -0
  253. /msprobe/{pytorch/doc → docs}/img/BLOOM-7B_4.png +0 -0
  254. /msprobe/{pytorch/doc → docs}/img/GPT-3_1.png +0 -0
  255. /msprobe/{pytorch/doc → docs}/img/GPT-3_2.png +0 -0
  256. /msprobe/{pytorch/doc → docs}/img/GPT-3_3.png +0 -0
  257. /msprobe/{pytorch/doc → docs}/img/GPT-3_4.png +0 -0
  258. /msprobe/{pytorch/doc → docs}/img/GPT-3_5.png +0 -0
  259. /msprobe/{pytorch/doc → docs}/img/GPT-3_6.png +0 -0
  260. /msprobe/{pytorch/doc → docs}/img/GPT-3_7.png +0 -0
  261. /msprobe/{pytorch/doc → docs}/img/GPT-3_8.png +0 -0
  262. /msprobe/{pytorch/doc → docs}/img/YOLOV5S_1.png +0 -0
  263. /msprobe/{pytorch/doc → docs}/img/YOLOV5S_2.png +0 -0
  264. /msprobe/{pytorch/doc → docs}/img/accuracy_checking_details.png +0 -0
  265. /msprobe/{pytorch/doc → docs}/img/accuracy_checking_result.png +0 -0
  266. /msprobe/{pytorch/doc → docs}/img/api_precision_compare_details.png +0 -0
  267. /msprobe/{pytorch/doc → docs}/img/api_precision_compare_result.png +0 -0
  268. /msprobe/{pytorch/doc → docs}/img/auto_analyze_log.png +0 -0
  269. /msprobe/{pytorch/doc → docs}/img/compare_result_pkl.png +0 -0
  270. /msprobe/{pytorch/doc → docs}/img/compare_result_pkl_md5.png.png +0 -0
  271. /msprobe/{pytorch/doc → docs}/img/cpu_info.png +0 -0
  272. /msprobe/{config → docs}/img/free_benchmark.png +0 -0
  273. /msprobe/{doc/grad_probe/img/image-1.png → docs/img/grad_probe_image-1.png} +0 -0
  274. /msprobe/{doc/grad_probe/img/image-2.png → docs/img/grad_probe_image-2.png} +0 -0
  275. /msprobe/{doc/grad_probe/img/image-3.png → docs/img/grad_probe_image-3.png} +0 -0
  276. /msprobe/{doc/grad_probe/img/image-4.png → docs/img/grad_probe_image-4.png} +0 -0
  277. /msprobe/{doc/grad_probe/img/image.png → docs/img/grad_probe_image.png} +0 -0
  278. /msprobe/{pytorch/doc → docs}/img/module_compare.png +0 -0
@@ -1,144 +1,153 @@
1
- import os
2
-
3
- from msprobe.core.data_dump.scope import build_scope, ListScope
4
- from msprobe.core.data_dump.json_writer import DataWriter
5
- from msprobe.core.common.log import logger
6
- from msprobe.core.common.const import Const
7
- from msprobe.core.data_dump.data_processor.factory import DataProcessorFactory
8
-
9
-
10
- def build_data_collector(config):
11
- return DataCollector(config)
12
-
13
-
14
- class DataCollector:
15
- multi_output_apis = ["_sort_", "npu_flash_attention"]
16
- tasks_need_tensor_data = [Const.OVERFLOW_CHECK, Const.TENSOR, Const.FREE_BENCHMARK]
17
- level_without_construct = ["L1", "L2"]
18
-
19
- def __init__(self, config):
20
- self.config = config
21
- self.data_writer = DataWriter()
22
- self.data_processor = DataProcessorFactory.create_processor(self.config, self.data_writer)
23
- self.module_processor = DataProcessorFactory.get_module_processor(self.config.framework) \
24
- if self.config.framework == Const.PT_FRAMEWORK else None
25
- self.module_count = {}
26
- if self.config.task == Const.FREE_BENCHMARK:
27
- self.scope = build_scope(ListScope, self.config.scope, self.config.list)
28
- else:
29
- self.scope = build_scope(None, self.config.scope, self.config.list)
30
-
31
- @property
32
- def dump_data_dir(self):
33
- return self.data_writer.dump_tensor_data_dir
34
-
35
- @property
36
- def dump_file_path(self):
37
- return self.data_writer.dump_file_path
38
-
39
- @staticmethod
40
- def check_scope_and_pid(scope, name, pid):
41
- return (not scope or scope.check(name)) and pid == os.getpid()
42
-
43
- @staticmethod
44
- def is_inplace(module):
45
- return getattr(module, "op_is_inplace", False)
46
-
47
- def if_return_forward_new_output(self):
48
- return self.data_processor.if_return_forward_new_output()
49
-
50
- def get_forward_new_output(self):
51
- return self.data_processor.get_forward_new_output()
52
-
53
- def visit_and_clear_overflow_status(self, api_or_module_name):
54
- self.data_processor.visit_and_clear_overflow_status(api_or_module_name)
55
-
56
- def write_json(self):
57
- self.data_writer.write_json()
58
-
59
- def update_data(self, data_info, msg=''):
60
- if self.config.task == Const.OVERFLOW_CHECK:
61
- if self.data_processor.has_overflow:
62
- self.data_writer.update_data(data_info)
63
- msg += "Overflow detected."
64
- else:
65
- msg += "No Overflow, OK."
66
- else:
67
- self.data_writer.update_data(data_info)
68
- return msg
69
-
70
- def pre_forward_data_collect(self, name, module, pid, module_input_output):
71
- backward_name = name.replace(Const.FORWARD, Const.BACKWARD)
72
- if self.check_scope_and_pid(self.scope, backward_name, pid):
73
- self.data_processor.analyze_pre_forward(backward_name, module, module_input_output)
74
- if not self.is_inplace(module) or not self.check_scope_and_pid(self.scope, name, pid):
75
- return
76
- logger.info(f"API {name} is inplace.")
77
- data_info = self.data_processor.analyze_pre_forward_inplace(name, module_input_output)
78
- self.handle_data(name, data_info)
79
-
80
- def forward_data_collect(self, name, module, pid, module_input_output):
81
- self.update_construct(name)
82
- if not self.check_scope_and_pid(self.scope, name, pid):
83
- return
84
-
85
- if not self.is_inplace(module):
86
- data_info = self.data_processor.analyze_forward(name, module, module_input_output)
87
- else:
88
- data_info = self.data_processor.analyze_forward_inplace(name, module_input_output)
89
- if self.config.level == "L2":
90
- return
91
- self.data_writer.update_stack(self.data_processor.analyze_api_call_stack(name))
92
- if self.data_processor.is_terminated:
93
- self.handle_data(name, data_info, use_buffer=False)
94
- raise Exception("[msprobe] exit")
95
- self.handle_data(name, data_info)
96
-
97
- def backward_data_collect(self, name, module, pid, module_input_output):
98
- self.update_construct(name)
99
- if not self.check_scope_and_pid(self.scope, name, pid):
100
- return
101
-
102
- data_info = self.data_processor.analyze_backward(name, module, module_input_output)
103
- if self.data_processor.is_terminated:
104
- self.handle_data(name, data_info, use_buffer=False)
105
- raise Exception("[msprobe] exit")
106
- self.handle_data(name, data_info)
107
-
108
- def backward_input_data_collect(self, name, module, pid, module_input_output):
109
- self.update_construct(name)
110
- if not self.check_scope_and_pid(self.scope, name, pid):
111
- return
112
-
113
- data_info = self.data_processor.analyze_backward_input(name, module, module_input_output)
114
- self.handle_data(name, data_info)
115
-
116
- def backward_output_data_collect(self, name, module, pid, module_input_output):
117
- self.update_construct(name)
118
- if not self.check_scope_and_pid(self.scope, name, pid):
119
- return
120
-
121
- data_info = self.data_processor.analyze_backward_output(name, module, module_input_output)
122
- self.handle_data(name, data_info)
123
-
124
- def update_construct(self, name):
125
- if self.config.framework == Const.PT_FRAMEWORK and self.config.level not in DataCollector.level_without_construct:
126
- self.data_writer.update_construct({name: self.module_processor.api_parent_node})
127
- self.data_writer.update_construct(self.module_processor.module_node)
128
-
129
- def handle_data(self, name, data_info, use_buffer=True):
130
- if data_info:
131
- msg = f"msprobe is collecting data on {name}. "
132
- msg = self.update_data(data_info, msg)
133
- logger.info(msg)
134
- if use_buffer:
135
- self.data_writer.flush_data_when_buffer_is_full()
136
- else:
137
- self.write_json()
138
-
139
- def update_dump_paths(self, *args):
140
- self.data_writer.update_dump_paths(*args)
141
- self.data_writer.initialize_json_file(task=self.config.task, level=self.config.level)
142
-
143
- def update_iter(self, current_iter):
144
- self.data_processor.update_iter(current_iter)
1
+ # Copyright (c) 2024-2024, Huawei Technologies Co., Ltd.
2
+ # All rights reserved.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ import os
17
+
18
+ from msprobe.core.data_dump.scope import build_scope, ListScope
19
+ from msprobe.core.data_dump.json_writer import DataWriter
20
+ from msprobe.core.common.log import logger
21
+ from msprobe.core.common.const import Const
22
+ from msprobe.core.data_dump.data_processor.factory import DataProcessorFactory
23
+
24
+
25
+ def build_data_collector(config):
26
+ return DataCollector(config)
27
+
28
+
29
+ class DataCollector:
30
+ multi_output_apis = ["_sort_", "npu_flash_attention"]
31
+ tasks_need_tensor_data = [Const.OVERFLOW_CHECK, Const.TENSOR, Const.FREE_BENCHMARK]
32
+ level_without_construct = [Const.LEVEL_L1, Const.LEVEL_L2]
33
+
34
+ def __init__(self, config):
35
+ self.config = config
36
+ self.data_writer = DataWriter()
37
+ self.data_processor = DataProcessorFactory.create_processor(self.config, self.data_writer)
38
+ self.module_processor = DataProcessorFactory.get_module_processor(self.config.framework)
39
+ self.module_count = {}
40
+ if self.config.task == Const.FREE_BENCHMARK:
41
+ self.scope = build_scope(ListScope, self.config.scope, self.config.list)
42
+ else:
43
+ self.scope = build_scope(None, self.config.scope, self.config.list)
44
+
45
+ def __del__(self):
46
+ self.write_json()
47
+
48
+ @property
49
+ def dump_data_dir(self):
50
+ return self.data_writer.dump_tensor_data_dir
51
+
52
+ @property
53
+ def dump_file_path(self):
54
+ return self.data_writer.dump_file_path
55
+
56
+ @staticmethod
57
+ def check_scope_and_pid(scope, name, pid):
58
+ return (not scope or scope.check(name)) and pid == os.getpid()
59
+
60
+ @staticmethod
61
+ def is_inplace(module):
62
+ return getattr(module, "op_is_inplace", False)
63
+
64
+ def if_return_forward_new_output(self):
65
+ return self.data_processor.if_return_forward_new_output()
66
+
67
+ def get_forward_new_output(self):
68
+ return self.data_processor.get_forward_new_output()
69
+
70
+ def update_api_or_module_name(self, api_or_module_name):
71
+ self.data_processor.update_api_or_module_name(api_or_module_name)
72
+
73
+ def write_json(self):
74
+ self.data_writer.write_json()
75
+
76
+ def update_data(self, name, data_info):
77
+ msg = f"msprobe is collecting data on {name}."
78
+ if self.config.task == Const.OVERFLOW_CHECK:
79
+ if self.data_processor.has_overflow:
80
+ msg += " Overflow detected."
81
+ logger.warning(msg)
82
+ self.data_writer.update_data(data_info)
83
+ return
84
+ logger.debug(msg)
85
+ self.data_writer.update_data(data_info)
86
+
87
+ def pre_forward_data_collect(self, name, module, pid, module_input_output):
88
+ backward_name = name.replace(Const.FORWARD, Const.BACKWARD)
89
+ if self.check_scope_and_pid(self.scope, backward_name, pid):
90
+ self.data_processor.analyze_pre_forward(backward_name, module, module_input_output)
91
+ if not self.is_inplace(module) or not self.check_scope_and_pid(self.scope, name, pid):
92
+ return
93
+ logger.info(f"API {name} is inplace.")
94
+ data_info = self.data_processor.analyze_pre_forward_inplace(name, module_input_output)
95
+ self.handle_data(name, data_info, flush=self.data_processor.is_terminated)
96
+
97
+ def forward_data_collect(self, name, module, pid, module_input_output):
98
+ self.update_construct(name)
99
+ if not self.check_scope_and_pid(self.scope, name, pid):
100
+ return
101
+
102
+ if not self.is_inplace(module):
103
+ data_info = self.data_processor.analyze_forward(name, module, module_input_output)
104
+ else:
105
+ data_info = self.data_processor.analyze_forward_inplace(name, module_input_output)
106
+ if self.config.level == "L2":
107
+ return
108
+ self.data_writer.update_stack(self.data_processor.analyze_api_call_stack(name))
109
+ self.handle_data(name, data_info, flush=self.data_processor.is_terminated)
110
+
111
+ def backward_data_collect(self, name, module, pid, module_input_output):
112
+ self.update_construct(name)
113
+ if not self.check_scope_and_pid(self.scope, name, pid):
114
+ return
115
+
116
+ data_info = self.data_processor.analyze_backward(name, module, module_input_output)
117
+ self.handle_data(name, data_info, flush=self.data_processor.is_terminated)
118
+
119
+ def backward_input_data_collect(self, name, module, pid, module_input_output):
120
+ self.update_construct(name)
121
+ if not self.check_scope_and_pid(self.scope, name, pid):
122
+ return
123
+
124
+ data_info = self.data_processor.analyze_backward_input(name, module, module_input_output)
125
+ self.handle_data(name, data_info)
126
+
127
+ def backward_output_data_collect(self, name, module, pid, module_input_output):
128
+ self.update_construct(name)
129
+ if not self.check_scope_and_pid(self.scope, name, pid):
130
+ return
131
+
132
+ data_info = self.data_processor.analyze_backward_output(name, module, module_input_output)
133
+ self.handle_data(name, data_info)
134
+
135
+ def update_construct(self, name):
136
+ if self.config.level not in DataCollector.level_without_construct:
137
+ self.data_writer.update_construct({name: self.module_processor.api_parent_node})
138
+ self.data_writer.update_construct(self.module_processor.module_node)
139
+
140
+ def handle_data(self, name, data_info, flush=False):
141
+ if data_info:
142
+ self.update_data(name, data_info)
143
+ if not flush:
144
+ self.data_writer.flush_data_periodically()
145
+ else:
146
+ self.write_json()
147
+
148
+ def update_dump_paths(self, *args):
149
+ self.data_writer.update_dump_paths(*args)
150
+ self.data_writer.initialize_json_file(task=self.config.task, level=self.config.level)
151
+
152
+ def update_iter(self, current_iter):
153
+ self.data_processor.update_iter(current_iter)