pytorch
pytorch copied to clipboard
Tensors and Dynamic neural networks in Python with strong GPU acceleration
Needs https://github.com/ROCm/pytorch/pull/1506 to be merged first * Add `.ci/docker/aotriton_version.txt` for details of aotriton tarball * Add logic to `cmake/External/aotriton.cmake` to use preinstalled aotriton if it exists * Update logic in...
Includes https://github.com/ROCm/triton/commit/5a02332983ee03f6727573120155d5ceb6c6376a
Includes https://github.com/ROCm/triton/commit/606262ac4b8d368b4f66f0789a884e1c4635c973 Tested via: http://ml-ci-internal.amd.com:8080/job/pytorch/job/dev/job/manylinux_rocm_wheels_test/136/ and http://ml-ci-internal.amd.com:8080/job/pytorch/job/dev/job/manylinux_rocm_wheels_test/140
Includes https://github.com/ROCm/triton/commit/6fe1cd7868145b4fad19ac8f5ae1a71e7774a139 Being tested via: http://ml-ci-internal.amd.com:8080/job/pytorch/job/dev/job/manylinux_rocm_wheels_test/143
Includes https://github.com/ROCm/triton/commit/888a431d740812a46842055361ada564db798f1f Tested via: http://ml-ci-internal.amd.com:8080/job/pytorch/job/dev/job/manylinux_rocm_wheels_test/134/ and http://ml-ci-internal.amd.com:8080/job/pytorch/job/dev/job/manylinux_rocm_wheels_test/138. Triton wheels built successfully
Fixes inductor.test_torchinductor_dynamic_shapes::TestInductorDynamicCUDA::test_item_unbacked_stride_nobreak_cuda
* Fail earlier for distributed-on-1-GPU scenario * print cmd in consolidated log with prettier formatting * python->python3 Fixes https://ontrack-internal.amd.com/browse/SWDEV-477264