Ting Lu comments

Results 49 comments of


                                            Ting Lu

CUDA 12.4 ARM wheel integration to CI

@pytorchbot rebase

CUDA 12.4 ARM wheel integration to CI

@pytorchbot merge -i

CUDA 12.4 ARM wheel integration to CI

@pytorchbot merge

Bump `nn.functional.conv3d` tolerances for `test_comprehensive`

@pytorchbot merge

Bump `nn.functional.conv3d` tolerances for `test_comprehensive`

@pytorchbot rebase

Bump `nn.functional.conv3d` tolerances for `test_comprehensive`

@pytorchbot merge

[CI] Add CUDA 13.0 inductor CI benchmarks

From cusparse doc, https://docs.nvidia.com/cuda/pdf/CUSPARSE_Library.pdf, suggested passing `-DDISABLE_CUSPARSE_DEPRECATED` to the compiler to suppress deprecation warnings. Trying.

[CI] Add CUDA 13.0 inductor CI benchmarks

Previous flag worked in resolving the deprecation warning. New build error ``` 2025-10-17T22:16:31.7451807Z [31m [0m [71/411] Building CUDA object CMakeFiles/fbgemm_gpu_py.dir/src/memory_utils/memory_utils.cu.o 2025-10-17T22:16:31.7452879Z [31m [0m FAILED: CMakeFiles/fbgemm_gpu_py.dir/src/memory_utils/memory_utils.cu.o 2025-10-17T22:16:31.7461457Z [31m [0m /usr/local/cuda-13.0/bin/nvcc -forward-unknown-to-host-compiler...

[CI] Add CUDA 13.0 inductor CI benchmarks

Disabled fbgemm for CUDA 13 inductor build due to https://github.com/pytorch/pytorch/pull/165029#issuecomment-3423107039. CUDA 13 support for FBGEMM is WIP https://github.com/pytorch/FBGEMM/pull/4783

[CI] Add CUDA 13.0 inductor CI benchmarks

Further debug shows that the job calls into install_torchrec_and_fbgemm function from https://github.com/pytorch/pytorch/blob/d049ed2cb1619c44279cb716b8a1d94e4df3b372/.ci/pytorch/test.sh#L1771, based on log "Building wheels for collected packages: torchrec". Therefore need to disable it from test file as...