mslk-cuda-nightly 2026.1.19__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.
Files changed (116) hide show
  1. mslk/__init__.py +56 -0
  2. mslk/attention/__init__.py +7 -0
  3. mslk/attention/cutlass_blackwell_fmha/__init__.py +30 -0
  4. mslk/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_custom_op.py +332 -0
  5. mslk/attention/cutlass_blackwell_fmha/cutlass_blackwell_fmha_interface.py +533 -0
  6. mslk/attention/flash_attn/__init__.py +22 -0
  7. mslk/attention/flash_attn/ampere_helpers.py +104 -0
  8. mslk/attention/flash_attn/barrier.py +72 -0
  9. mslk/attention/flash_attn/benchmark.py +269 -0
  10. mslk/attention/flash_attn/blackwell_helpers.py +754 -0
  11. mslk/attention/flash_attn/block_info.py +109 -0
  12. mslk/attention/flash_attn/block_sparse_utils.py +1452 -0
  13. mslk/attention/flash_attn/block_sparsity.py +219 -0
  14. mslk/attention/flash_attn/compute_block_sparsity.py +378 -0
  15. mslk/attention/flash_attn/copy_utils.py +341 -0
  16. mslk/attention/flash_attn/cute_dsl_utils.py +135 -0
  17. mslk/attention/flash_attn/fast_math.py +22 -0
  18. mslk/attention/flash_attn/flash_bwd.py +1262 -0
  19. mslk/attention/flash_attn/flash_bwd_postprocess.py +464 -0
  20. mslk/attention/flash_attn/flash_bwd_preprocess.py +366 -0
  21. mslk/attention/flash_attn/flash_bwd_sm100.py +2951 -0
  22. mslk/attention/flash_attn/flash_bwd_sm90.py +1703 -0
  23. mslk/attention/flash_attn/flash_fwd.py +2471 -0
  24. mslk/attention/flash_attn/flash_fwd_combine.py +705 -0
  25. mslk/attention/flash_attn/flash_fwd_sm100.py +2727 -0
  26. mslk/attention/flash_attn/hopper_helpers.py +102 -0
  27. mslk/attention/flash_attn/interface.py +1771 -0
  28. mslk/attention/flash_attn/mask.py +610 -0
  29. mslk/attention/flash_attn/mma_sm100_desc.py +292 -0
  30. mslk/attention/flash_attn/named_barrier.py +32 -0
  31. mslk/attention/flash_attn/pack_gqa.py +165 -0
  32. mslk/attention/flash_attn/paged_kv.py +176 -0
  33. mslk/attention/flash_attn/pipeline.py +273 -0
  34. mslk/attention/flash_attn/seqlen_info.py +139 -0
  35. mslk/attention/flash_attn/softmax.py +583 -0
  36. mslk/attention/flash_attn/testing.py +424 -0
  37. mslk/attention/flash_attn/tile_scheduler.py +720 -0
  38. mslk/attention/flash_attn/utils.py +860 -0
  39. mslk/attention/fmha/__init__.py +967 -0
  40. mslk/attention/fmha/_triton/__init__.py +6 -0
  41. mslk/attention/fmha/_triton/available.py +50 -0
  42. mslk/attention/fmha/_triton/splitk_kernels.py +1534 -0
  43. mslk/attention/fmha/_triton/vararg_kernel.py +262 -0
  44. mslk/attention/fmha/attn_bias.py +2186 -0
  45. mslk/attention/fmha/attn_bias_utils.py +536 -0
  46. mslk/attention/fmha/ck.py +508 -0
  47. mslk/attention/fmha/ck_decoder.py +141 -0
  48. mslk/attention/fmha/ck_splitk.py +204 -0
  49. mslk/attention/fmha/common.py +598 -0
  50. mslk/attention/fmha/cutlass.py +461 -0
  51. mslk/attention/fmha/cutlass_blackwell.py +560 -0
  52. mslk/attention/fmha/dispatch.py +224 -0
  53. mslk/attention/fmha/flash.py +862 -0
  54. mslk/attention/fmha/flash3.py +858 -0
  55. mslk/attention/fmha/flash_mtia.py +245 -0
  56. mslk/attention/fmha/merge_training.py +192 -0
  57. mslk/attention/fmha/split_blocks_fairinternal.py +329 -0
  58. mslk/attention/fmha/torch_attention_compat.py +154 -0
  59. mslk/attention/fmha/tree_attention.py +718 -0
  60. mslk/attention/fmha/triton_splitk.py +1378 -0
  61. mslk/attention/fmha/unbind.py +130 -0
  62. mslk/attention/fmha/utils/__init__.py +6 -0
  63. mslk/attention/fmha/utils/bench.py +74 -0
  64. mslk/attention/fmha/utils/cpp_lib.py +148 -0
  65. mslk/attention/fmha/utils/op_common.py +65 -0
  66. mslk/attention/gqa_attn_splitk/__init__.py +11 -0
  67. mslk/bench/comm/__init__.py +7 -0
  68. mslk/bench/comm/comm_bench.py +255 -0
  69. mslk/bench/common/__init__.py +5 -0
  70. mslk/bench/common/utils.py +148 -0
  71. mslk/bench/conv/__init__.py +7 -0
  72. mslk/bench/conv/conv_bench.py +551 -0
  73. mslk/bench/conv/conv_ops.py +213 -0
  74. mslk/bench/gemm/__init__.py +7 -0
  75. mslk/bench/gemm/gemm_bench.py +859 -0
  76. mslk/bench/gemm/gemm_ops.py +3342 -0
  77. mslk/bench/gemm/grouped_gemm_bias_scale_benchmark.py +177 -0
  78. mslk/bench/moe/__init__.py +7 -0
  79. mslk/bench/moe/gather_scatter_bench.py +356 -0
  80. mslk/bench/quantize/quantize_bench.py +345 -0
  81. mslk/bench/quantize/quantize_ops.py +266 -0
  82. mslk/comm/__init__.py +11 -0
  83. mslk/conv/__init__.py +11 -0
  84. mslk/gemm/__init__.py +18 -0
  85. mslk/gemm/triton/__init__.py +7 -0
  86. mslk/gemm/triton/fp8_gemm.py +2702 -0
  87. mslk/gemm/triton/grouped_gemm.py +1132 -0
  88. mslk/gemm/triton/matmul_perf_model.py +237 -0
  89. mslk/gemm/triton/utils.py +128 -0
  90. mslk/kv_cache/__init__.py +11 -0
  91. mslk/moe/__init__.py +26 -0
  92. mslk/moe/activation.py +291 -0
  93. mslk/moe/gather_scatter.py +739 -0
  94. mslk/moe/layers.py +1240 -0
  95. mslk/moe/shuffling.py +421 -0
  96. mslk/mslk.so +0 -0
  97. mslk/quantize/__init__.py +11 -0
  98. mslk/quantize/shuffle.py +306 -0
  99. mslk/quantize/triton/__init__.py +7 -0
  100. mslk/quantize/triton/fp4_quantize.py +5942 -0
  101. mslk/quantize/triton/fp8_quantize.py +1902 -0
  102. mslk/testing/__init__.py +7 -0
  103. mslk/testing/attributes.py +60 -0
  104. mslk/testing/rocm.py +91 -0
  105. mslk/utils/__init__.py +7 -0
  106. mslk/utils/torch/__init__.py +7 -0
  107. mslk/utils/torch/library.py +150 -0
  108. mslk/utils/triton/__init__.py +7 -0
  109. mslk/utils/triton/fp8_utils.py +72 -0
  110. mslk/utils/triton/utils.py +128 -0
  111. mslk/version.py +11 -0
  112. mslk_cuda_nightly-2026.1.19.dist-info/METADATA +102 -0
  113. mslk_cuda_nightly-2026.1.19.dist-info/RECORD +116 -0
  114. mslk_cuda_nightly-2026.1.19.dist-info/WHEEL +5 -0
  115. mslk_cuda_nightly-2026.1.19.dist-info/licenses/LICENSE +30 -0
  116. mslk_cuda_nightly-2026.1.19.dist-info/top_level.txt +1 -0
@@ -0,0 +1,6 @@
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
+ # pyre-unsafe
@@ -0,0 +1,50 @@
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
+ # pyre-unsafe
7
+
8
+ import logging
9
+ import os
10
+
11
+ import torch
12
+
13
+
14
+ logger = logging.getLogger("xformers")
15
+
16
+
17
+ def compute_once(func):
18
+ value = None
19
+
20
+ def func_wrapper():
21
+ nonlocal value
22
+ if value is None:
23
+ value = func()
24
+ return value
25
+
26
+ return func_wrapper
27
+
28
+
29
+ @compute_once
30
+ def is_triton_available():
31
+ if os.environ.get("XFORMERS_ENABLE_TRITON", "0") == "1":
32
+ return True
33
+ if not torch.cuda.is_available():
34
+ return False
35
+ if os.environ.get("XFORMERS_FORCE_DISABLE_TRITON", "0") == "1":
36
+ return False
37
+ # We have many errors on V100 with recent triton versions
38
+ # Let's just drop support for triton kernels below A100
39
+ if torch.cuda.get_device_capability("cuda") < (8, 0):
40
+ return False
41
+ try:
42
+ import triton # noqa
43
+
44
+ return True
45
+ except (ImportError, AttributeError):
46
+ logger.warning(
47
+ "A matching Triton is not available, some optimizations will not be enabled",
48
+ exc_info=True,
49
+ )
50
+ return False