Commits · 80c5079cd5058ab5b5af456ad27c892e9f825f82 · OpenDAS / TransformerEngine

23 May, 2025 1 commit
- [DCU] surpport blockwise int8 quant · 80c5079c
  yuguo authored May 23, 2025
  
  80c5079c
22 May, 2025 3 commits
- Fix build error of userbuffer.cu · c636071d
  wenjh authored May 22, 2025
```
Signed-off-by: wenjh <wenjh@sugon.com>
```
  c636071d
- [DCU] Add width to __shfl · 6ed9a3e4
  wenjh authored May 22, 2025
```
Signed-off-by: wenjh <wenjh@sugon.com>
```
  6ed9a3e4
- [ROCM6.3] Fix build on rocm-6.3 · b27e513d
  wenjh authored May 22, 2025
```
Signed-off-by: wenjh <wenjh@sugon.com>
```
  b27e513d
21 May, 2025 1 commit
- [DCU] remove cudaStreamSynchronize for tp overlap · 92d59fe4
  yuguo authored May 21, 2025
  
  92d59fe4
20 May, 2025 3 commits
- [DCU] cudaStreamSynchronize for tp gemm overlap · aec86199
  yuguo authored May 20, 2025
  
  aec86199
- [DCU] surpport delay_wgrad_compute in batchgemm · 460b006c
  yuguo authored May 20, 2025
  
  460b006c
- [DCU] variable ub streams add NVTE_UB_STREAM_NUMS · 196a213f
  yuguo authored May 20, 2025
  
  196a213f
14 May, 2025 1 commit

[HIPIFY] Add rules for hipify · a916fb7b

wenjh authored May 14, 2025



Add rules of cuda_runtime.h, cuda_driver.h and cuda_nvml.h to hip.
Signed-off-by: wenjh <wenjh@sugon.com>

a916fb7b

13 May, 2025 1 commit
- [DCU] surpport blockwise fp8 quantize · b8fe26e7
  yuguo authored May 13, 2025
  
  b8fe26e7
08 May, 2025 2 commits

[DCU] add batchgemm test · 9d0f1c9b
yuguo authored May 08, 2025

9d0f1c9b

[ROCBLAS_GEMM] Default use of hipMallocAsync · 7a47930f

wenjh authored May 08, 2025



Default use of hipMallocAsync rather than hipMalloc in rocblas_gemm and
add support of fp16_fp16_fp32 in rocblas_gemm.
Signed-off-by: wenjh <wenjh@sugon.com>

7a47930f

07 May, 2025 2 commits
- [DCU] fix batchgemm · e8f92b93
  yuguo authored May 07, 2025
  
  e8f92b93
- [DCU] surpport NVTE_MOE_BATCHCOUNT · c37084b9
  yuguo authored May 07, 2025
  
  c37084b9
06 May, 2025 4 commits
- [DCU] fix batch gemm · c686efc1
  yuguo authored May 06, 2025
  
  c686efc1
- [DCU] Fix launch bounds · cdb862cd
  wenjh authored May 06, 2025
```
Fix launch bounds of multi_tensor_apply_kernel and
thd_out_correction_kernel.
Signed-off-by: wenjh <wenjh@sugon.com>
```
  cdb862cd
- [DCU] new rocm gemm · 229be5e8
  yuguo authored May 06, 2025
  
  229be5e8
- [ROCM_GEMM] Fix launch params · 6efebcd0
  wenjh authored May 06, 2025
```
Fix launch params are larger than launch bounds(256) for kernels in
rocm_gemm.cu
Signed-off-by: wenjh <wenjh@sugon.com>
```
  6efebcd0
30 Apr, 2025 1 commit

[rocblas] Use HandleManager to avoid mem leakage · 388ac735

wenjh authored Apr 30, 2025


Signed-off-by: wenjh <wenjh@sugon.com>

[RocblasGemm] Provide support of AB(bf16)D(fp32)
Signed-off-by: wenjh <wenjh@sugon.com>

388ac735

29 Apr, 2025 3 commits
- Fix missing import · 08f14085
  wenjh authored Apr 29, 2025
  
  08f14085
- [DCU] Fix WS leak when init+destroy ub more than 1 · 8fda607c
  yuguo authored Apr 29, 2025
  
  8fda607c
- [PytorchUnitTest] Fix errors while running tests · 86f2e9a9
  wenjh authored Apr 29, 2025
```
Signed-off-by: wenjh <wenjh@sugon.com>
```
  86f2e9a9
27 Apr, 2025 1 commit

[DTK-25.04] Workaround compiler bugs. · 5749aec6

wenjh authored Apr 27, 2025



Ref params of rmsnorm will make program corruption with 'nil' error.
Signed-off-by: wenjh <wenjh@sugon.com>

5749aec6

25 Apr, 2025 3 commits
- [DCU] tp overlap opt · caf2fbf2
  yuguo authored Apr 25, 2025
  
  caf2fbf2
- [DAS-RMSNorm] TE 2.3 returns 3 values of rmsnorm · daa15293
  panning authored Apr 25, 2025
```
API `rmsnorm_forward` of python returns 3 values rather than 2 from V2.3
Signed-off-by: wenjh <wenjh@sugon.com>
```
  daa15293
- [DCU] fix rocblas backend · e80f260d
  yuguo authored Apr 25, 2025
  
  e80f260d
24 Apr, 2025 2 commits

[DCU] Fix failed test cases · 3ce226ae

wenjh authored Apr 23, 2025



Due to the difference of warp size between nvidia(32) and dtk(64), the
OperatorTest/CTDBiasTestSuite.TestCTDBias/* are all failed except:

* OperatorTest/CTDBiasTestSuite.TestCTDBias/bfloat16Xfloat32X65536X128
* OperatorTest/CTDBiasTestSuite.TestCTDBias/bfloat16Xfloat16X65536X128
* OperatorTest/CTDBiasTestSuite.TestCTDBias/bfloat16Xbfloat16X65536X128
* OperatorTest/CTDBiasTestSuite.TestCTDBias/bfloat16Xfloat8e5m2X65536X128
* OperatorTest/CTDBiasTestSuite.TestCTDBias/bfloat16Xfloat8e4m3X65536X128

This commit is intended to fix this.
Signed-off-by: wenjh <wenjh@sugon.com>

3ce226ae

[DCU] Fix crash test cases · 46c81675

wenjh authored Apr 19, 2025



Due to the compiler compiling incorrect code. The following test case crashed:

* OperatorTest/CTTestSuite.TestCastTranspose/bfloat16Xbfloat16X2048X12288
* OperatorTest/CTTestSuite.TestCastTranspose/bfloat16Xbfloat16X65536X128
* OperatorTest/CTTestSuite.TestCastTranspose/bfloat16Xbfloat16X256X65536

This commit is intended to fix these test cases.
Signed-off-by: wenjh <wenjh@sugon.com>

46c81675

23 Apr, 2025 1 commit
- [DCU] surpport rocm gemm rocblas · 8b27a2b7
  yuguo authored Apr 23, 2025
  
  8b27a2b7
22 Apr, 2025 1 commit
- [DCU] little fix · 73f3ac47
  yuguo authored Apr 22, 2025
  
  73f3ac47
18 Apr, 2025 3 commits

[DCU] overlap bug fix in ECO and BW finally · 456a96c8
yuguo authored Apr 18, 2025

456a96c8

Split wgrad&dgrad from backward() to support a2a overlap (#1653) · 9f8aaddf

Hongbin Liu authored Apr 18, 2025



* split wgrad for GroupedLinear
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* support wgrad split for linear and ln_linear
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* add comments and fix WeightGradStore
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* support bias and fix unit tests
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* minor fix
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* support fuse_grad_accumulation=false
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* add wgrad split for layernorm_mlp
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* minor fix
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* fix unittest
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* add unittest for distributed interface apply Dener's suggestion
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* minor fix
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* replace split_bw with delay_wgrad_compute
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* Update transformer_engine/pytorch/module/layernorm_mlp.py
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Update transformer_engine/pytorch/module/linear.py
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* Update transformer_engine/pytorch/module/layernorm_linear.py
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

* remove comments
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>

---------
Signed-off-by: Hongbin Liu <hongbinl@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: Hongbin Liu <hongbinl@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

9f8aaddf

[JAX] Deprecate Praxis layers (#1694) · 1a6a6d7b

Phuong Nguyen authored Apr 17, 2025



rm pax/praxis
Signed-off-by: Phuong Nguyen <phuonguyen@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

1a6a6d7b

17 Apr, 2025 6 commits

Re Do symmetric memory merge request (#1682) · 39c0e709

wdykas authored Apr 17, 2025



* re merge request
Signed-off-by: Peter Dykas <wdykas@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* add docstring
Signed-off-by: Peter Dykas <wdykas@nvidia.com>

---------
Signed-off-by: Peter Dykas <wdykas@nvidia.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

39c0e709

[PyTorch] Move swizzle scaling factor to cpp (#1683) · 4e036c8c

Xin Yao authored Apr 18, 2025



* move swizzle scaling factor to cpp
Signed-off-by: Xin Yao <xiny@nvidia.com>

* resolve comments
Signed-off-by: Xin Yao <xiny@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



---------
Signed-off-by: Xin Yao <xiny@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

4e036c8c

Allow NVTEShape to own data. (#1674) · e61ce77c

kwyss-nvidia authored Apr 17, 2025



* Allow NVTEShape to own data.
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

* Convert repeated copy paths to nvte_make_shape calls.
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

* Apply suggestions from code review
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

* Build fixes.
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

* MR feedback.
Signed-off-by: Keith Wyss <kwyss@nvidia.com>

---------
Signed-off-by: Keith Wyss <kwyss@nvidia.com>
Signed-off-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

e61ce77c

Support computing zero-centered gamma in compute dtype for CuDNN (#1690) · 61f1bf6f

jberchtold-nvidia authored Apr 17, 2025



* Add a flag to support computing zero-centered gamma in weight dtype or compute dtype for CuDNN
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

* Address comments
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

---------
Signed-off-by: Jeremy Berchtold <jberchtold@nvidia.com>

61f1bf6f

[DCU] tmp fix overlap test · b9ec4909
yuguo authored Apr 17, 2025

b9ec4909

[PyTorch] Deprecate the weight offloading (#1678) · 61312d6a

Paweł Gadziński authored Apr 17, 2025



* drop
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

---------
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>
Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>

61312d6a

16 Apr, 2025 1 commit

[Pytorch] NVIDIA-DL-Framework-Inspect support – part 1 – core (#1614) · beaecf84

Paweł Gadziński authored Apr 16, 2025



* add
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* weight workspace fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* docs fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* file i forgot
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* lint fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* Update transformer_engine/debug/pytorch/utils.py
Co-authored-by: Przemyslaw Tredak <ptrendx@gmail.com>
Signed-off-by: Paweł Gadziński <62263673+pggPL@users.noreply.github.com>

* setup fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* setup fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* Update transformer_engine/pytorch/tensor/_internal/float8_tensor_base.py
Co-authored-by: Przemyslaw Tredak <ptrendx@gmail.com>
Signed-off-by: Paweł Gadziński <62263673+pggPL@users.noreply.github.com>

* all tensor types
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fixes
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fixes
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fixes
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* removed check
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* move error
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* _reset
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* Update transformer_engine/pytorch/module/linear.py
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Signed-off-by: Paweł Gadziński <62263673+pggPL@users.noreply.github.com>

* name documentation
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* added blockwise quantizer
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



* make debug option optional
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

* Update transformer_engine/pytorch/tensor/quantized_tensor.py
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>
Signed-off-by: Paweł Gadziński <62263673+pggPL@users.noreply.github.com>

* names fix
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>

---------
Signed-off-by: Pawel Gadzinski <pgadzinski@nvidia.com>
Signed-off-by: Paweł Gadziński <62263673+pggPL@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Przemyslaw Tredak <ptrendx@gmail.com>
Co-authored-by: Tim Moon <4406448+timmoon10@users.noreply.github.com>

beaecf84