Commit Graph

7 Commits

Author SHA1 Message Date
Tom Ritchford
d25e6e623f Fix unused Python variables in test/[a-d]* (#134665)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134665
Approved by: https://github.com/albanD
2024-12-13 22:13:12 +00:00
Arun Pa
f71e368969 UFMT formatting on test/autograd test/ao test/cpp test/backends (#123369)
Partially addresses #123062

Ran lintrunner on
- test/_test_bazel.py
- test/ao
- test/autograd test/backends test/benchmark_uitls test/conftest.py test/bottleneck_test test/cpp

Pull Request resolved: https://github.com/pytorch/pytorch/pull/123369
Approved by: https://github.com/huydhn
2024-04-05 18:51:38 +00:00
Salil Desai
dc6ce1485e Use Variable Size Indices in Sparse Qlinear Code (#85247)
Final changes to enable sparse weight packing with variable size indices

pack_block_sparse.cc is deleted because all functions in it have a template added, so they are moved to pack_block_sparse.h

Differential Revision: [D39025651](https://our.internmc.facebook.com/intern/diff/D39025651/)

**NOTE FOR REVIEWERS**: This PR has internal Meta-specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D39025651/)!
Pull Request resolved: https://github.com/pytorch/pytorch/pull/85247
Approved by: https://github.com/digantdesai
2022-10-13 05:50:04 +00:00
Salil Desai
facbddb9ff Override Quantized Backend to use Fbgemm in Qlinear Packed Params Test (#86236)
Summary: After D39934051, we must explicitly ```override_quantized_engine('fbgemm')``` for this test to work

Test Plan:
```
buck test //caffe2/test:ao -- TestQlinearPackedParams
```

Before:
```
Started reporting to test run: https://www.internalfb.com/intern/testinfra/testrun/5629499663624574
    ✓ ListingSuccess: caffe2/test:ao : 72 tests discovered (32.830)
    ✓ Pass: caffe2/test:ao - test_qlinear_packed_params_qnnpack (ao.sparsity.test_qlinear_packed_params.TestQlinearPackedParams) (25.085)
    ✗ Fail: caffe2/test:ao - test_qlinear_packed_params (ao.sparsity.test_qlinear_packed_params.TestQlinearPackedParams) (26.706)
Test output:
> RuntimeError: Didn't find engine for operation ao::sparse::qlinear_prepack X86
```

After:
```
Started reporting to test run: https://www.internalfb.com/intern/testinfra/testrun/7599824485968786
    ✓ ListingSuccess: caffe2/test:ao : 72 tests discovered (31.082)
    ✓ Pass: caffe2/test:ao - test_qlinear_packed_params_fbgemm (ao.sparsity.test_qlinear_packed_params.TestQlinearPackedParams) (100.409)
    ✓ Pass: caffe2/test:ao - test_qlinear_packed_params_qnnpack (ao.sparsity.test_qlinear_packed_params.TestQlinearPackedParams) (100.544)
Summary
  Pass: 2
  ListingSuccess: 1
```

Differential Revision: D40078176

Pull Request resolved: https://github.com/pytorch/pytorch/pull/86236
Approved by: https://github.com/jmdetloff, https://github.com/z-a-f
2022-10-07 11:58:41 +00:00
Salil Desai
98e4524dcc [PyTorch Edge] Add OwnedOrBorrowedVector for QNNPack BCSR Indices/Values (#80476)
OwnedOrBorrowedVector allows us to create a BCSR which uses preexisting data pointers for its indices and values, so we can avoid copying data when loading. It also supports creating a BCSR which owns indices and values data in vectors as before.

Differential Revision: [D36956640](https://our.internmc.facebook.com/intern/diff/D36956640/)

**NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D36956640/)!
Pull Request resolved: https://github.com/pytorch/pytorch/pull/80476
Approved by: https://github.com/qihqi
2022-07-07 15:34:32 +00:00
Salil Desai
5c12cd224f [PyTorch Edge] Add qnnpack bcsr matrix unpacking and use unpacking in Linear module (#80475)
Having unpacking removes the need to store the original dense weights in the python Linear module

Differential Revision: [D34699287](https://our.internmc.facebook.com/intern/diff/D34699287/)

**NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D34699287/)!
Pull Request resolved: https://github.com/pytorch/pytorch/pull/80475
Approved by: https://github.com/qihqi
2022-07-07 15:32:21 +00:00
Salil Desai
eaf817df3a [PyTorch Edge] Add serialization/deserialization of Sparse Quantize Linear Packed Params (#80474)
Packed Params are serialized/deserialized in sparse form

Differential Revision: [D34392761](https://our.internmc.facebook.com/intern/diff/D34392761/)

**NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D34392761/)!
Pull Request resolved: https://github.com/pytorch/pytorch/pull/80474
Approved by: https://github.com/qihqi
2022-07-07 15:30:02 +00:00