fbgemm-gpu-genai-nightly 2025.12.19__cp310-cp310-manylinux_2_28_x86_64.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.

Potentially problematic release.

This version of fbgemm-gpu-genai-nightly might be problematic. Click here for more details.

Files changed (127) hide show
  1. fbgemm_gpu/__init__.py +186 -0
  2. fbgemm_gpu/asmjit.so +0 -0
  3. fbgemm_gpu/batched_unary_embeddings_ops.py +87 -0
  4. fbgemm_gpu/config/__init__.py +9 -0
  5. fbgemm_gpu/config/feature_list.py +88 -0
  6. fbgemm_gpu/docs/__init__.py +18 -0
  7. fbgemm_gpu/docs/common.py +9 -0
  8. fbgemm_gpu/docs/examples.py +73 -0
  9. fbgemm_gpu/docs/jagged_tensor_ops.py +259 -0
  10. fbgemm_gpu/docs/merge_pooled_embedding_ops.py +36 -0
  11. fbgemm_gpu/docs/permute_pooled_embedding_ops.py +108 -0
  12. fbgemm_gpu/docs/quantize_ops.py +41 -0
  13. fbgemm_gpu/docs/sparse_ops.py +616 -0
  14. fbgemm_gpu/docs/target.genai.json.py +6 -0
  15. fbgemm_gpu/enums.py +24 -0
  16. fbgemm_gpu/experimental/example/__init__.py +29 -0
  17. fbgemm_gpu/experimental/example/fbgemm_gpu_experimental_example_py.so +0 -0
  18. fbgemm_gpu/experimental/example/utils.py +20 -0
  19. fbgemm_gpu/experimental/gemm/triton_gemm/__init__.py +15 -0
  20. fbgemm_gpu/experimental/gemm/triton_gemm/fp4_quantize.py +5654 -0
  21. fbgemm_gpu/experimental/gemm/triton_gemm/fp8_gemm.py +4422 -0
  22. fbgemm_gpu/experimental/gemm/triton_gemm/grouped_gemm.py +1192 -0
  23. fbgemm_gpu/experimental/gemm/triton_gemm/matmul_perf_model.py +232 -0
  24. fbgemm_gpu/experimental/gemm/triton_gemm/utils.py +130 -0
  25. fbgemm_gpu/experimental/gen_ai/__init__.py +56 -0
  26. fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/__init__.py +46 -0
  27. fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_custom_op.py +333 -0
  28. fbgemm_gpu/experimental/gen_ai/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_interface.py +552 -0
  29. fbgemm_gpu/experimental/gen_ai/bench/__init__.py +13 -0
  30. fbgemm_gpu/experimental/gen_ai/bench/comm_bench.py +257 -0
  31. fbgemm_gpu/experimental/gen_ai/bench/gather_scatter_bench.py +348 -0
  32. fbgemm_gpu/experimental/gen_ai/bench/quantize_bench.py +707 -0
  33. fbgemm_gpu/experimental/gen_ai/bench/quantize_ops.py +3483 -0
  34. fbgemm_gpu/experimental/gen_ai/fbgemm_gpu_experimental_gen_ai.so +0 -0
  35. fbgemm_gpu/experimental/gen_ai/moe/README.md +15 -0
  36. fbgemm_gpu/experimental/gen_ai/moe/__init__.py +66 -0
  37. fbgemm_gpu/experimental/gen_ai/moe/activation.py +292 -0
  38. fbgemm_gpu/experimental/gen_ai/moe/gather_scatter.py +740 -0
  39. fbgemm_gpu/experimental/gen_ai/moe/layers.py +1272 -0
  40. fbgemm_gpu/experimental/gen_ai/moe/shuffling.py +421 -0
  41. fbgemm_gpu/experimental/gen_ai/quantize.py +307 -0
  42. fbgemm_gpu/fbgemm.so +0 -0
  43. fbgemm_gpu/metrics.py +160 -0
  44. fbgemm_gpu/permute_pooled_embedding_modules.py +142 -0
  45. fbgemm_gpu/permute_pooled_embedding_modules_split.py +85 -0
  46. fbgemm_gpu/quantize/__init__.py +43 -0
  47. fbgemm_gpu/quantize/quantize_ops.py +64 -0
  48. fbgemm_gpu/quantize_comm.py +315 -0
  49. fbgemm_gpu/quantize_utils.py +246 -0
  50. fbgemm_gpu/runtime_monitor.py +237 -0
  51. fbgemm_gpu/sll/__init__.py +189 -0
  52. fbgemm_gpu/sll/cpu/__init__.py +80 -0
  53. fbgemm_gpu/sll/cpu/cpu_sll.py +1001 -0
  54. fbgemm_gpu/sll/meta/__init__.py +35 -0
  55. fbgemm_gpu/sll/meta/meta_sll.py +337 -0
  56. fbgemm_gpu/sll/triton/__init__.py +127 -0
  57. fbgemm_gpu/sll/triton/common.py +38 -0
  58. fbgemm_gpu/sll/triton/triton_dense_jagged_cat_jagged_out.py +72 -0
  59. fbgemm_gpu/sll/triton/triton_jagged2_to_padded_dense.py +221 -0
  60. fbgemm_gpu/sll/triton/triton_jagged_bmm.py +418 -0
  61. fbgemm_gpu/sll/triton/triton_jagged_bmm_jagged_out.py +553 -0
  62. fbgemm_gpu/sll/triton/triton_jagged_dense_elementwise_add.py +52 -0
  63. fbgemm_gpu/sll/triton/triton_jagged_dense_elementwise_mul_jagged_out.py +175 -0
  64. fbgemm_gpu/sll/triton/triton_jagged_dense_flash_attention.py +861 -0
  65. fbgemm_gpu/sll/triton/triton_jagged_flash_attention_basic.py +667 -0
  66. fbgemm_gpu/sll/triton/triton_jagged_self_substraction_jagged_out.py +73 -0
  67. fbgemm_gpu/sll/triton/triton_jagged_softmax.py +463 -0
  68. fbgemm_gpu/sll/triton/triton_multi_head_jagged_flash_attention.py +751 -0
  69. fbgemm_gpu/sparse_ops.py +1455 -0
  70. fbgemm_gpu/split_embedding_configs.py +452 -0
  71. fbgemm_gpu/split_embedding_inference_converter.py +175 -0
  72. fbgemm_gpu/split_embedding_optimizer_ops.py +21 -0
  73. fbgemm_gpu/split_embedding_utils.py +29 -0
  74. fbgemm_gpu/split_table_batched_embeddings_ops.py +73 -0
  75. fbgemm_gpu/split_table_batched_embeddings_ops_common.py +484 -0
  76. fbgemm_gpu/split_table_batched_embeddings_ops_inference.py +2042 -0
  77. fbgemm_gpu/split_table_batched_embeddings_ops_training.py +4600 -0
  78. fbgemm_gpu/split_table_batched_embeddings_ops_training_common.py +146 -0
  79. fbgemm_gpu/ssd_split_table_batched_embeddings_ops.py +26 -0
  80. fbgemm_gpu/tbe/__init__.py +6 -0
  81. fbgemm_gpu/tbe/bench/__init__.py +55 -0
  82. fbgemm_gpu/tbe/bench/bench_config.py +156 -0
  83. fbgemm_gpu/tbe/bench/bench_runs.py +709 -0
  84. fbgemm_gpu/tbe/bench/benchmark_click_interface.py +187 -0
  85. fbgemm_gpu/tbe/bench/eeg_cli.py +137 -0
  86. fbgemm_gpu/tbe/bench/embedding_ops_common_config.py +149 -0
  87. fbgemm_gpu/tbe/bench/eval_compression.py +119 -0
  88. fbgemm_gpu/tbe/bench/reporter.py +35 -0
  89. fbgemm_gpu/tbe/bench/tbe_data_config.py +137 -0
  90. fbgemm_gpu/tbe/bench/tbe_data_config_bench_helper.py +323 -0
  91. fbgemm_gpu/tbe/bench/tbe_data_config_loader.py +289 -0
  92. fbgemm_gpu/tbe/bench/tbe_data_config_param_models.py +170 -0
  93. fbgemm_gpu/tbe/bench/utils.py +48 -0
  94. fbgemm_gpu/tbe/cache/__init__.py +11 -0
  95. fbgemm_gpu/tbe/cache/kv_embedding_ops_inference.py +385 -0
  96. fbgemm_gpu/tbe/cache/split_embeddings_cache_ops.py +48 -0
  97. fbgemm_gpu/tbe/ssd/__init__.py +15 -0
  98. fbgemm_gpu/tbe/ssd/common.py +46 -0
  99. fbgemm_gpu/tbe/ssd/inference.py +586 -0
  100. fbgemm_gpu/tbe/ssd/training.py +4908 -0
  101. fbgemm_gpu/tbe/ssd/utils/__init__.py +7 -0
  102. fbgemm_gpu/tbe/ssd/utils/partially_materialized_tensor.py +273 -0
  103. fbgemm_gpu/tbe/stats/__init__.py +10 -0
  104. fbgemm_gpu/tbe/stats/bench_params_reporter.py +339 -0
  105. fbgemm_gpu/tbe/utils/__init__.py +13 -0
  106. fbgemm_gpu/tbe/utils/common.py +42 -0
  107. fbgemm_gpu/tbe/utils/offsets.py +65 -0
  108. fbgemm_gpu/tbe/utils/quantize.py +251 -0
  109. fbgemm_gpu/tbe/utils/requests.py +556 -0
  110. fbgemm_gpu/tbe_input_multiplexer.py +108 -0
  111. fbgemm_gpu/triton/__init__.py +22 -0
  112. fbgemm_gpu/triton/common.py +77 -0
  113. fbgemm_gpu/triton/jagged/__init__.py +8 -0
  114. fbgemm_gpu/triton/jagged/triton_jagged_tensor_ops.py +824 -0
  115. fbgemm_gpu/triton/quantize.py +647 -0
  116. fbgemm_gpu/triton/quantize_ref.py +286 -0
  117. fbgemm_gpu/utils/__init__.py +11 -0
  118. fbgemm_gpu/utils/filestore.py +211 -0
  119. fbgemm_gpu/utils/loader.py +36 -0
  120. fbgemm_gpu/utils/torch_library.py +132 -0
  121. fbgemm_gpu/uvm.py +40 -0
  122. fbgemm_gpu_genai_nightly-2025.12.19.dist-info/METADATA +62 -0
  123. fbgemm_gpu_genai_nightly-2025.12.19.dist-info/RECORD +127 -0
  124. fbgemm_gpu_genai_nightly-2025.12.19.dist-info/WHEEL +5 -0
  125. fbgemm_gpu_genai_nightly-2025.12.19.dist-info/top_level.txt +2 -0
  126. list_versions/__init__.py +12 -0
  127. list_versions/cli_run.py +163 -0
fbgemm_gpu/experimental/example/utils.py ADDED
@@ -0,0 +1,20 @@
1
+ #!/usr/bin/env python3
2
+ # Copyright (c) Meta Platforms, Inc. and affiliates.
3
+ # All rights reserved.
4
+ #
5
+ # This source code is licensed under the BSD-style license found in the
6
+ # LICENSE file in the root directory of this source tree.
7
+
8
+ # pyre-strict
9
+
10
+ import torch
11
+
12
+
13
+ def add_tensors(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
14
+ return torch.ops.fbgemm.add_tensors_float(a, b)
15
+
16
+
17
+ def sgemm(
18
+ alpha: float, TA: torch.Tensor, TB: torch.Tensor, beta: float, TC: torch.Tensor
19
+ ) -> torch.Tensor:
20
+ return torch.ops.fbgemm.sgemm_float(alpha, TA, TB, beta, TC)
fbgemm_gpu/experimental/gemm/triton_gemm/__init__.py ADDED
@@ -0,0 +1,15 @@
1
+ #!/usr/bin/env python3
2
+ # Copyright (c) Meta Platforms, Inc. and affiliates.
3
+ # All rights reserved.
4
+ #
5
+ # This source code is licensed under the BSD-style license found in the
6
+ # LICENSE file in the root directory of this source tree.
7
+
8
+ # pyre-strict
9
+
10
+ try:
11
+ # pyre-ignore[21]
12
+ # @manual=//deeplearning/fbgemm/fbgemm_gpu:test_utils
13
+ from fbgemm_gpu import open_source
14
+ except Exception:
15
+ open_source: bool = False