blksprs 2.0rc6__py3-none-any.whl → 2.0rc8__py3-none-any.whl

blksprs/utils/tools.py

@@ -1,3 +1,6 @@
+import tomllib
+from pathlib import Path
+
 import torch
 from torch import Tensor, Size
 
@@ -5,6 +8,11 @@ from torch import Tensor, Size
 torch._dynamo.config.capture_scalar_outputs = True
 
 
+def version():
+    with open(Path(__file__).parent.parent.parent.joinpath("pyproject.toml"), "rb") as f:
+        return tomllib.load(f)["project"]["version"]
+
+
 def do_shape_blocksparse(x: Tensor):
     if x.dim() == 3:
         return x.contiguous(), x.size()
@@ -26,12 +34,3 @@ def stride(x: Tensor):
         return x.size(1) * x.size(2), x.size(2), 1
     else:
         raise NotImplementedError
-
-
-def get_autocast_min_val():
-    if torch.is_autocast_enabled():
-        dtype = torch.get_autocast_dtype("cuda")
-    else:
-        dtype = torch.float
-
-    return torch.finfo(dtype).min

{blksprs-2.0rc6.dist-info → blksprs-2.0rc8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: blksprs
-Version: 2.0rc6
+Version: 2.0rc8
 Summary: A lightweight library for operations on blocksparse matrices in PyTorch.
 Author-email: Felix Schön <schoen@kr.tuwien.ac.at>
 Project-URL: Homepage, https://github.com/FelixSchoen/blksprs
@@ -108,12 +108,16 @@ library.
 
 ## Known Limitations and Issues
 
+- Triton has a bug with `tl.atomix_max()` used for the row-wise max operation.
+  In order to work around this bug a manual conversion of some values is needed, (slightly) negatively impacting
+  performance.
+  Watch the [issue](https://github.com/triton-lang/triton/issues/6376) on Triton's issue tracker for more information.
 - PyTorch's `wrap_triton()` currently does not support config pruning. It thus cannot be used for some of the kernels,
   which could impact graph compilation.
 - There seem to be some issues with autocasting, forcing some operations to manually cast.
 - There will be some slight numerical differences between vanilla and blksprs operations.
-These instabilities are due to Triton and thus cannot be fixed by this library alone.
-However, for all intents and purposes, these very minor differences should not matter and can safely be ignored.
+  These instabilities are due to Triton and thus cannot be fixed by this library alone.
+  However, for all intents and purposes, these very minor differences should not matter and can safely be ignored.
 
 ## Usage
 

blksprs-2.0rc8.dist-info/RECORD

@@ -0,0 +1,23 @@
+blksprs/__init__.py,sha256=283rF0fbrUqsH_KXUvCgbCMqO0GOgenMkxwDVh1QdpU,1617
+blksprs/layouting/distribution_layout.py,sha256=ur1ty_2U-Hfj78hMWsLZvu7ZuGhzW3qGLKMc72DfTZM,5861
+blksprs/layouting/sparsity_layout.py,sha256=eXHmu2h7K5Q-YUpfOxocJoeP_5ZoQFZf_eHLxRZQbYU,11207
+blksprs/ops/conversion.py,sha256=_g32aEEZdeuHHPj1pBfTNMxknRwJ9O1zk3Wv76pBIrg,21898
+blksprs/ops/distribution.py,sha256=0tPldv0ARzmCV1CU2jvfqpHBgOuHPrDFiCtqsLs7CZc,20789
+blksprs/ops/flow.py,sha256=PDZAD8u4y9qW1IXERki6ItKbEKnm_ChG8SKWM3_P9Oc,8245
+blksprs/ops/matmul.py,sha256=5tVBKU_lglUjaLDi6J_dscdqlmzRz38OGxqAxZxZXDs,11879
+blksprs/ops/partitioning.py,sha256=cfQmY9BZqGTvvJorIhtb-EyuGRJGPraWR-wTKdb47aI,9954
+blksprs/ops/repeat.py,sha256=TLYNxwPuT9y5K9xyM41WK5gnggAJF3lI61Q2K7zWjns,9035
+blksprs/ops/softmax.py,sha256=BwrRQdtRdkiSvl2mf5bpsTmyIxWiJOpa1HFg0st5yGU,12778
+blksprs/ops/transpose.py,sha256=U-VAyLRT6_NDv9qYSFzBqfVlDeIpTqAMEXkqto0VF6w,4072
+blksprs/ops/misc/broadcast_ops.py,sha256=-PrHiSJikZh8nXUmXxSCtFEP27TTxFr4wcrNxBjnimk,5987
+blksprs/ops/misc/row_wise.py,sha256=n5FJjAuOd8BHBJQx4bsQwr-HmXkR9PYVAqfk77wjOFU,19653
+blksprs/utils/autotuning.py,sha256=a-kmWRjJ3eED2XbjkQeOJSyW8bdIs27HgKMPvAKqWeU,2052
+blksprs/utils/benchmarking.py,sha256=dLabDscTFn5NkmOI1g7DnKeTneUYW3RIVv9MDF-8BKc,1271
+blksprs/utils/blksprs_tensor.py,sha256=pfoz59aJixj_fIoFx76ySiygwRQUemmgjMKepZ2c4j0,244
+blksprs/utils/processing.py,sha256=RNkEDc0g-sNHRuMPkRzNWU13d3_lIkXMJdoqES4yQTM,3738
+blksprs/utils/tools.py,sha256=BozpH3oEXe3K9ZRJsIzlasDk-sZyJqmwSf1gl7xbbdo,865
+blksprs/utils/validation.py,sha256=G8eQlvJVMKfEX3k2AwBD0A6Ck-gFoRLpLNY6HXsB3fA,4348
+blksprs-2.0rc8.dist-info/METADATA,sha256=h70L26BthR6laP7sMQLF9L3dHIRQNCF_oKwZ5g4dZSg,9509
+blksprs-2.0rc8.dist-info/WHEEL,sha256=zaaOINJESkSfm_4HQVc5ssNzHCPXhJm0kEUakpsEHaU,91
+blksprs-2.0rc8.dist-info/top_level.txt,sha256=qyp0IHeY3H2GQA97i4hk_To5rRBS2YcE1HRPSLy04fk,8
+blksprs-2.0rc8.dist-info/RECORD,,

{blksprs-2.0rc6.dist-info → blksprs-2.0rc8.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (78.1.0)
+Generator: setuptools (80.8.0)
 Root-Is-Purelib: true
 Tag: py3-none-any
 

blksprs-2.0rc6.dist-info/RECORD

@@ -1,23 +0,0 @@
-blksprs/__init__.py,sha256=OHfpwJCZWGUfpT-DVfC1YSaeZl4aCMNt9CrzMPymywU,1577
-blksprs/layouting/distribution_layout.py,sha256=TkMh_DYKX56Cb8Vq7EHyupMRvzm0XbUNP8QP7afv9wM,5122
-blksprs/layouting/sparsity_layout.py,sha256=6GOjwllDUK9L8jEQNu2i17Pp1BIIQm8fv3xVuiR0zIw,10228
-blksprs/ops/conversion.py,sha256=2zAdbaZ1iP2lisLVeG-k-f571G4HJapADhSwpY0Zd3o,21503
-blksprs/ops/distribution.py,sha256=Gffhd7z85IDm57G8x_v7J2P3ezEVOHEWPD_36Lf8Irs,20453
-blksprs/ops/flow.py,sha256=UO5ba5TFgVpEyT7r0hnWYw3vhRDpBOxyPHUBeNOAYPs,7935
-blksprs/ops/matmul.py,sha256=02hujXMtFgF7ohepM3v6h9okrfcU-J3mQZV17B-qvh0,12235
-blksprs/ops/partitioning.py,sha256=nAV28f3NtvT4OFvDtnE0A-VxpDQmMXS0pZw4CJwzqGA,9838
-blksprs/ops/repeat.py,sha256=bQpJuwtt8aRdSzxT78lJ8f8fLDhPkYK5UvMfJ-PQrkc,8977
-blksprs/ops/softmax.py,sha256=-NoTf1Cpuku9C99N0LuMydT_ObozWTnZJGDZxseXEXI,12209
-blksprs/ops/transpose.py,sha256=PQKteFnzNAOEC7voO7wh_dq9c54UjCboJz889aBCwKc,4010
-blksprs/ops/misc/broadcast_ops.py,sha256=wBusOtscfGSbtfsCynI1ypr93KuCfVpLX_4b4l8-dck,5811
-blksprs/ops/misc/row_wise.py,sha256=k23p1rizOLS_iRWFhiKiRW6KnR2qxmHfsE8jq0VFfa0,18991
-blksprs/utils/autotuning.py,sha256=tDfMWklm2rvbo0-ahH81C3Gg0U6LHjPn3d_3pEOzmJs,2053
-blksprs/utils/benchmarking.py,sha256=dLabDscTFn5NkmOI1g7DnKeTneUYW3RIVv9MDF-8BKc,1271
-blksprs/utils/blksprs_tensor.py,sha256=pfoz59aJixj_fIoFx76ySiygwRQUemmgjMKepZ2c4j0,244
-blksprs/utils/processing.py,sha256=xuu9iDpwTvsqI_WKMSD8QCNuvPnfcKMRcuF2L4Zs6Ts,3808
-blksprs/utils/tools.py,sha256=NusV0H_XPn4ETJTibQwh3bJBqfW12iUrBuk1EfjbAQs,851
-blksprs/utils/validation.py,sha256=G8eQlvJVMKfEX3k2AwBD0A6Ck-gFoRLpLNY6HXsB3fA,4348
-blksprs-2.0rc6.dist-info/METADATA,sha256=gC-91T17-byW4_f0R41hWjwzOr-N1fjp-HIDu3phunU,9179
-blksprs-2.0rc6.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
-blksprs-2.0rc6.dist-info/top_level.txt,sha256=qyp0IHeY3H2GQA97i4hk_To5rRBS2YcE1HRPSLy04fk,8
-blksprs-2.0rc6.dist-info/RECORD,,