From 13e472a3bca6121446e9911e3b6f1970f9a64f90 Mon Sep 17 00:00:00 2001 From: Prachi Gupta Date: Tue, 5 Aug 2025 15:48:46 +0000 Subject: [PATCH 1/8] Fix issues with merge conflicts - Fix the tensorpipe branch for ROCm - External/aotriton.cmake: remove use of __AOTRITON_VER_WITH_COMMIT - macros/Export.h: remove TORCH_HIP_CPP_API/TORCH_HIP_API as CUDA ones get hipified and converted correctly (need to upstream this) - CUDALoops.cuh: Bad merge - Blas.cpp: remove MX patch - cuda_vectorized_test.cu: remove ROCmloops specific test, this was removed in rocm7.0_internal_testing branch. I had incorrectly addressed the merge conflicts when merging with upstream --- .gitmodules | 3 +- aten/src/ATen/native/cuda/Blas.cpp | 21 ++---------- aten/src/ATen/native/cuda/CUDALoops.cuh | 38 ---------------------- aten/src/ATen/test/cuda_vectorized_test.cu | 17 ---------- cmake/External/aotriton.cmake | 2 +- third_party/tensorpipe | 2 +- torch/headeronly/macros/Export.h | 16 +++------ 7 files changed, 10 insertions(+), 89 deletions(-) diff --git a/.gitmodules b/.gitmodules index 88a3fd091a881..74a0b360a8af0 100644 --- a/.gitmodules +++ b/.gitmodules @@ -81,7 +81,8 @@ [submodule "third_party/tensorpipe"] ignore = dirty path = third_party/tensorpipe - url = https://github.com/pytorch/tensorpipe.git + url = https://github.com/ROCm/tensorpipe.git + branch = tp_rocm_60 [submodule "third_party/cudnn_frontend"] path = third_party/cudnn_frontend url = https://github.com/NVIDIA/cudnn-frontend.git diff --git a/aten/src/ATen/native/cuda/Blas.cpp b/aten/src/ATen/native/cuda/Blas.cpp index ef144e9d99c38..40d39b3c7b606 100644 --- a/aten/src/ATen/native/cuda/Blas.cpp +++ b/aten/src/ATen/native/cuda/Blas.cpp @@ -1205,7 +1205,6 @@ std::pair get_joint_scaling( } // namespace - // Computes matrix multiply + bias while applying scaling to input and output matrices // Scales are only applicable when matrices are of Float8 type and assumed to be equal to 1.0 by default. // If output matrix type is 16 or 32-bit type, scale_result is not applied. @@ -1362,25 +1361,9 @@ _scaled_mm_out_cuda(const Tensor& mat1, const Tensor& mat2, else { TORCH_CHECK(b.dtype() == at::kFloat8_e4m3fn); } - // Until more than bf16 is supported + // Until more than bf16 is supported. TORCH_CHECK(out.scalar_type() == ScalarType::BFloat16, - "hipblaslt rowwise _scaled_mm only supports BFloat16 output"); - } - else if (scaling_choice == ScalingType::BlockWise) { -#if ROCM_VERSION >= 70000 - TORCH_CHECK(at::detail::getCUDAHooks().isGPUArch({"gfx950"}), - "Block-wise scaling for Float8_e8m0fnu is only supported on gfx950"); - - TORCH_CHECK(mat1.size(0) % 32 == 0 && mat1.size(1) % 32 == 0 && - mat2.size(0) % 32 == 0 && mat2.size(1) % 32 == 0, - "Matrix dimensions must be multiples of 32 for block-wise scaling"); - - TORCH_CHECK(out.scalar_type() == ScalarType::BFloat16 || - out.scalar_type() == ScalarType::Half, - "Block-wise scaling only supports BFloat16 or Half output types"); -#else - TORCH_CHECK(false, "Block-wise scaling for Float8_e8m0fnu requires ROCm 7.0 or later"); -#endif + "hipblaslt rowwise _scaled_mm only supports BFloat16 output but got ", out.scalar_type()); } #endif diff --git a/aten/src/ATen/native/cuda/CUDALoops.cuh b/aten/src/ATen/native/cuda/CUDALoops.cuh index 3767fe1d9a88b..9b104a7966363 100644 --- a/aten/src/ATen/native/cuda/CUDALoops.cuh +++ b/aten/src/ATen/native/cuda/CUDALoops.cuh @@ -531,44 +531,6 @@ __global__ void elementwise_kernel(int N, func_t f) { } } -#ifdef USE_ROCM -template -C10_LAUNCH_BOUNDS_2(nt, 4) -__global__ void elementwise_kernel_manual_unroll(int N, func_t f) { - int tid = threadIdx.x; - int nv = nt * vt; - int idx = nv * blockIdx.x + tid; - if ((idx + nt*(vt-1)) < N) { - f(idx, true); - } else { -#pragma unroll - for (int i = 0; i < vt; i++) { - if (idx < N) { - f(idx, false); - idx += nt; - } - } - } -} - -template -C10_LAUNCH_BOUNDS_2(nt, 4) -__global__ void elementwise_kernel_strided(int N, func_t f) { - int tid = threadIdx.x; - int idx = nt * vt * blockIdx.x + tid; - int step = nt * vt * gridDim.x; - while (idx < N) { -#pragma unroll - for (int i = 0; i < vt; i++) { - if ((idx + nt * i) < N) { - f(idx + nt * i); - } - } - idx += step; - } -} -#endif - template static void launch_legacy_kernel(int64_t N, const func_t& f) { TORCH_INTERNAL_ASSERT(N >= 0 && N <= std::numeric_limits::max()); diff --git a/aten/src/ATen/test/cuda_vectorized_test.cu b/aten/src/ATen/test/cuda_vectorized_test.cu index 7ba7bcb99bce1..236753c94d37b 100644 --- a/aten/src/ATen/test/cuda_vectorized_test.cu +++ b/aten/src/ATen/test/cuda_vectorized_test.cu @@ -27,23 +27,6 @@ void reset_buffers() { } } -#if defined(USE_ROCM) && !defined(_WIN32) -TEST(TestLoops, HasSameArgTypes) { - // This is a compile-time unit test. If this file compiles without error, - // then the test passes and during runtime, we just need to return. - using namespace at::native::modern::detail; - using func1_t = int (*)(float, float); - using func2_t = int (*)(bool, float, float); - using func3_t = int (*)(float); - using func4_t = int (*)(); - static_assert(has_same_arg_types::value, "func1_t has the same argument types"); - static_assert(!has_same_arg_types::value, "func2_t does not have the same argument types"); - static_assert(has_same_arg_types::value, "func3_t has the same argument types"); - static_assert(has_same_arg_types::value, "func4_t has the same argument types"); - return; -} -#endif - TEST(TestVectorizedMemoryAccess, CanVectorizeUpTo) { char *ptr = reinterpret_cast(buffer1); diff --git a/cmake/External/aotriton.cmake b/cmake/External/aotriton.cmake index 172521a592c3a..54564e42c9023 100644 --- a/cmake/External/aotriton.cmake +++ b/cmake/External/aotriton.cmake @@ -81,7 +81,7 @@ if(NOT __AOTRITON_INCLUDED) list(GET __AOTRITON_MANYLINUX_LIST ${__AOTRITON_ROCM_INDEX} __AOTRITON_MANYLINUX) set(__AOTRITON_ARCH ${CMAKE_HOST_SYSTEM_PROCESSOR}) string(CONCAT __AOTRITON_FILE "aotriton-" - "${__AOTRITON_VER_WITH_COMMIT}-${__AOTRITON_MANYLINUX}" + "${__AOTRITON_VER}-${__AOTRITON_MANYLINUX}" "_${__AOTRITON_ARCH}-rocm${__AOTRITON_ROCM}" "-shared.tar.${__AOTRITON_Z}") string(CONCAT __AOTRITON_URL "https://github.com/ROCm/aotriton/releases/download/" # @lint-ignore diff --git a/third_party/tensorpipe b/third_party/tensorpipe index 135ba25f6be99..c1dec474c59ec 160000 --- a/third_party/tensorpipe +++ b/third_party/tensorpipe @@ -1 +1 @@ -Subproject commit 135ba25f6be9991ebfe83d41d268d9c3d4cc5c5b +Subproject commit c1dec474c59ec8cbb629cfe992977062de08b9bd diff --git a/torch/headeronly/macros/Export.h b/torch/headeronly/macros/Export.h index 8dd25419efb4e..a241439c86bca 100644 --- a/torch/headeronly/macros/Export.h +++ b/torch/headeronly/macros/Export.h @@ -100,10 +100,10 @@ #define TORCH_API C10_IMPORT #endif -// You may be wondering why we have TORCH_CUDA_CPP_API and TORCH_CUDA_CU_API +// You may be wondering why we have TORCH_HIP_CPP_API and TORCH_HIP_API // belonging to the same library instead of just one TORCH_CUDA_API. Well, it -// can indeed just be one TORCH_CUDA_API (and used to be)! TORCH_CUDA_CPP_API -// and TORCH_CUDA_CU_API are artifacts of when we needed a split build to +// can indeed just be one TORCH_CUDA_API (and used to be)! TORCH_HIP_CPP_API +// and TORCH_HIP_API are artifacts of when we needed a split build to // avoid relocation marker linking errors. The context is as follows: // // Once upon a time, there _was_ only TORCH_CUDA_API. All was happy until we @@ -130,14 +130,6 @@ #define TORCH_CUDA_CU_API C10_IMPORT #endif -#if defined(TORCH_HIP_BUILD_MAIN_LIB) -#define TORCH_HIP_CPP_API C10_EXPORT -#define TORCH_HIP_API C10_EXPORT -#else -#define TORCH_HIP_CPP_API C10_IMPORT -#define TORCH_HIP_API C10_IMPORT -#endif - #if defined(TORCH_XPU_BUILD_MAIN_LIB) #define TORCH_XPU_API C10_EXPORT #else @@ -145,7 +137,7 @@ #endif // Enums only need to be exported on windows for non-CUDA files -#if defined(_WIN32) && defined(__CUDACC__) +#if defined(_WIN32) && defined(__HIPCC__) #define C10_API_ENUM C10_API #else #define C10_API_ENUM From 7bd7047ad708ae240f538c3159fe2d5a5ab49a60 Mon Sep 17 00:00:00 2001 From: Prachi Gupta Date: Wed, 6 Aug 2025 20:27:29 +0000 Subject: [PATCH 2/8] Update triton commit and related_commits files - Update triton commit to point to ToT of release/internal/3.4.x - Update related_commits file with ToT of respective repos --- .ci/docker/ci_commit_pins/triton.txt | 2 +- related_commits | 22 ++++++++++------------ 2 files changed, 11 insertions(+), 13 deletions(-) diff --git a/.ci/docker/ci_commit_pins/triton.txt b/.ci/docker/ci_commit_pins/triton.txt index 60c896b80c8f4..567536db72100 100644 --- a/.ci/docker/ci_commit_pins/triton.txt +++ b/.ci/docker/ci_commit_pins/triton.txt @@ -1 +1 @@ -f7888497a1eb9e98d4c07537f0d0bcfe180d1363 +711e2a92522e0a9921ce58ae658571ca55c49b97 diff --git a/related_commits b/related_commits index 95ebcf476dbce..71aeaf72071b9 100644 --- a/related_commits +++ b/related_commits @@ -1,12 +1,10 @@ -ubuntu|pytorch|apex|master|ca54c058f1094b3463788371325025be707a5982|https://github.com/ROCm/apex -centos|pytorch|apex|master|ca54c058f1094b3463788371325025be707a5982|https://github.com/ROCm/apex -ubuntu|pytorch|torchvision|main|f52c4f1afd7dec25cbe7b98bcf1cbc840298e8da|https://github.com/pytorch/vision -centos|pytorch|torchvision|main|f52c4f1afd7dec25cbe7b98bcf1cbc840298e8da|https://github.com/pytorch/vision -ubuntu|pytorch|torchtext|main|bde7ecdb6ba9179ccd30cde60a6550478d0a359f|https://github.com/pytorch/text -centos|pytorch|torchtext|main|bde7ecdb6ba9179ccd30cde60a6550478d0a359f|https://github.com/pytorch/text -ubuntu|pytorch|torchdata|main|922ac065407546b9cb4f629ab99f1fbf04d8fc12|https://github.com/pytorch/data -centos|pytorch|torchdata|main|922ac065407546b9cb4f629ab99f1fbf04d8fc12|https://github.com/pytorch/data -ubuntu|pytorch|torchaudio|main|bccaa454a54c3c648697cc2f46a4fb0500b1f01b|https://github.com/pytorch/audio -centos|pytorch|torchaudio|main|bccaa454a54c3c648697cc2f46a4fb0500b1f01b|https://github.com/pytorch/audio -ubuntu|pytorch|ao|main|a96eeb1c7d7ba24cf0ccfc105141729acfed22bf|https://github.com/pytorch/ao -centos|pytorch|ao|main|a96eeb1c7d7ba24cf0ccfc105141729acfed22bf|https://github.com/pytorch/ao +ubuntu|pytorch|apex|master|62c94ed1789bc177a83567985be6c1cb29b2d98c|https://github.com/ROCm/apex +centos|pytorch|apex|master|62c94ed1789bc177a83567985be6c1cb29b2d98c|https://github.com/ROCm/apex +ubuntu|pytorch|torchvision|main|98f8b3757c0648724064ca95434b18281c43c5f6|https://github.com/pytorch/vision +centos|pytorch|torchvision|main|98f8b3757c0648724064ca95434b18281c43c5f6|https://github.com/pytorch/vision +ubuntu|pytorch|torchdata|main|a05a54f797dd0f1a66610652a949fd47243ff952|https://github.com/pytorch/data +centos|pytorch|torchdata|main|a05a54f797dd0f1a66610652a949fd47243ff952|https://github.com/pytorch/data +ubuntu|pytorch|torchaudio|main|0c22347335f4c9a5b92a2f5bad65e05e2464c184|https://github.com/pytorch/audio +centos|pytorch|torchaudio|main|0c22347335f4c9a5b92a2f5bad65e05e2464c184|https://github.com/pytorch/audio +ubuntu|pytorch|ao|main|3b4bc9869d933927b2547d8231feab69789a80d4|https://github.com/pytorch/ao +centos|pytorch|ao|main|3b4bc9869d933927b2547d8231feab69789a80d4|https://github.com/pytorch/ao From 9025071507f9ca4ba1f9bfecf9c7630e0100930b Mon Sep 17 00:00:00 2001 From: Prachi Gupta Date: Wed, 6 Aug 2025 21:28:44 +0000 Subject: [PATCH 3/8] Revert "CONSOLIDATED COMMITS: Enable tensorpipe with hip_basic backend" This reverts commit 550bc776de086c3cdd8103da42d90574fc764223. --- .gitmodules | 3 +-- cmake/Dependencies.cmake | 12 ++---------- third_party/tensorpipe | 2 +- torch/csrc/distributed/rpc/tensorpipe_cuda.cpp | 6 +----- torch/testing/_internal/distributed/rpc/rpc_test.py | 1 - 5 files changed, 5 insertions(+), 19 deletions(-) diff --git a/.gitmodules b/.gitmodules index 74a0b360a8af0..88a3fd091a881 100644 --- a/.gitmodules +++ b/.gitmodules @@ -81,8 +81,7 @@ [submodule "third_party/tensorpipe"] ignore = dirty path = third_party/tensorpipe - url = https://github.com/ROCm/tensorpipe.git - branch = tp_rocm_60 + url = https://github.com/pytorch/tensorpipe.git [submodule "third_party/cudnn_frontend"] path = third_party/cudnn_frontend url = https://github.com/NVIDIA/cudnn-frontend.git diff --git a/cmake/Dependencies.cmake b/cmake/Dependencies.cmake index a514924589a04..d11915fe43147 100644 --- a/cmake/Dependencies.cmake +++ b/cmake/Dependencies.cmake @@ -1159,14 +1159,6 @@ if(USE_DISTRIBUTED AND USE_TENSORPIPE) set(TP_USE_CUDA ON CACHE BOOL "" FORCE) set(TP_ENABLE_CUDA_IPC ON CACHE BOOL "" FORCE) endif() - if(USE_ROCM) - add_compile_options(-D__HIP_PLATFORM_AMD__=1) - set(TP_USE_ROCM ON CACHE BOOL "" FORCE) - set(TP_ENABLE_HIP_IPC OFF CACHE BOOL "" FORCE) - set(TP_ENABLE_HIP_XTH OFF CACHE BOOL "" FORCE) - set(TP_ENABLE_HIP_GDR OFF CACHE BOOL "" FORCE) - set(TP_ENABLE_IBV OFF CACHE BOOL "" FORCE) - endif() set(TP_BUILD_LIBUV ON CACHE BOOL "" FORCE) add_compile_options(-DTORCH_USE_LIBUV) include_directories(BEFORE SYSTEM ${CMAKE_CURRENT_LIST_DIR}/../third_party/tensorpipe/third_party/libuv/include) @@ -1192,9 +1184,9 @@ if(USE_DISTRIBUTED AND USE_TENSORPIPE) if(USE_CUDA) list(APPEND Caffe2_CUDA_DEPENDENCY_LIBS tensorpipe_cuda) elseif(USE_ROCM) - message(WARNING "TensorPipe is supported on ROCm") + message(WARNING "TensorPipe doesn't yet support ROCm") # Not yet... - list(APPEND Caffe2_HIP_DEPENDENCY_LIBS tensorpipe_hip) + # list(APPEND Caffe2_HIP_DEPENDENCY_LIBS tensorpipe_hip) endif() endif() endif() diff --git a/third_party/tensorpipe b/third_party/tensorpipe index c1dec474c59ec..dacda0567d9f2 160000 --- a/third_party/tensorpipe +++ b/third_party/tensorpipe @@ -1 +1 @@ -Subproject commit c1dec474c59ec8cbb629cfe992977062de08b9bd +Subproject commit dacda0567d9f23d4bc503e1c4f84aa65f33ac38a diff --git a/torch/csrc/distributed/rpc/tensorpipe_cuda.cpp b/torch/csrc/distributed/rpc/tensorpipe_cuda.cpp index 97fb18cac83bb..03b43184d143b 100644 --- a/torch/csrc/distributed/rpc/tensorpipe_cuda.cpp +++ b/torch/csrc/distributed/rpc/tensorpipe_cuda.cpp @@ -1,7 +1,7 @@ #include #include -#if defined(USE_TENSORPIPE) +#if defined(USE_TENSORPIPE) && !defined(USE_ROCM) #include #include @@ -50,8 +50,6 @@ C10_REGISTER_CREATOR(TensorPipeChannelRegistry, cuda_gdr, makeCudaGdrChannel) #endif -#if TENSORPIPE_HAS_CUDA_XTH_CHANNEL - std::unique_ptr makeCudaXthChannel() { auto context = tensorpipe::channel::cuda_xth::create(); return std::make_unique( @@ -61,8 +59,6 @@ std::unique_ptr makeCudaXthChannel() { // The cuda_xth channel supports same-process GPU-to-GPU comm C10_REGISTER_CREATOR(TensorPipeChannelRegistry, cuda_xth, makeCudaXthChannel) -#endif - std::unique_ptr makeCudaBasicChannel() { auto context = tensorpipe::channel::cuda_basic::create( tensorpipe::channel::basic::create()); diff --git a/torch/testing/_internal/distributed/rpc/rpc_test.py b/torch/testing/_internal/distributed/rpc/rpc_test.py index ea7c0d125c82c..631a32bb4fcdf 100644 --- a/torch/testing/_internal/distributed/rpc/rpc_test.py +++ b/torch/testing/_internal/distributed/rpc/rpc_test.py @@ -37,7 +37,6 @@ captured_output, skip_if_lt_x_gpu, tp_transports, - skip_if_rocm, ) from torch.testing._internal.common_utils import ( get_cycles_per_ms, From 301dc4677dd48bce4043953cb39e6f30db2a63de Mon Sep 17 00:00:00 2001 From: Prachi Gupta Date: Fri, 8 Aug 2025 16:05:41 +0000 Subject: [PATCH 4/8] Export.h: dehipify this file --- torch/headeronly/macros/Export.h | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/torch/headeronly/macros/Export.h b/torch/headeronly/macros/Export.h index a241439c86bca..2222baedb4b1b 100644 --- a/torch/headeronly/macros/Export.h +++ b/torch/headeronly/macros/Export.h @@ -100,10 +100,10 @@ #define TORCH_API C10_IMPORT #endif -// You may be wondering why we have TORCH_HIP_CPP_API and TORCH_HIP_API +// You may be wondering why we have TORCH_CUDA_CPP_API and TORCH_CUDA_API // belonging to the same library instead of just one TORCH_CUDA_API. Well, it -// can indeed just be one TORCH_CUDA_API (and used to be)! TORCH_HIP_CPP_API -// and TORCH_HIP_API are artifacts of when we needed a split build to +// can indeed just be one TORCH_CUDA_API (and used to be)! TORCH_CUDA_CPP_API +// and TORCH_CUDA_API are artifacts of when we needed a split build to // avoid relocation marker linking errors. The context is as follows: // // Once upon a time, there _was_ only TORCH_CUDA_API. All was happy until we @@ -137,7 +137,7 @@ #endif // Enums only need to be exported on windows for non-CUDA files -#if defined(_WIN32) && defined(__HIPCC__) +#if defined(_WIN32) && defined(__CUDACC__) #define C10_API_ENUM C10_API #else #define C10_API_ENUM From 9ce6b1a1254bbd7630eb5773d23bb36d631f5db7 Mon Sep 17 00:00:00 2001 From: Prachi Gupta Date: Fri, 8 Aug 2025 16:58:01 +0000 Subject: [PATCH 5/8] Update requirements-ci with upstream and rocm/release/2.8 --- .ci/docker/requirements-ci.txt | 31 ++++++++++++++----------------- 1 file changed, 14 insertions(+), 17 deletions(-) diff --git a/.ci/docker/requirements-ci.txt b/.ci/docker/requirements-ci.txt index feac690887a46..5cdcaa3d81a33 100644 --- a/.ci/docker/requirements-ci.txt +++ b/.ci/docker/requirements-ci.txt @@ -110,10 +110,8 @@ ninja==1.11.1.3 #Pinned versions: 1.11.1.3 #test that import: run_test.py, test_cpp_extensions_aot.py,test_determination.py -numba==0.49.0 ; python_version < "3.9" -numba==0.55.2 ; python_version == "3.9" -numba==0.55.2 ; python_version == "3.10" -numba==0.60.0 ; python_version == "3.12" +numba==0.60.0 ; python_version == "3.9" +numba==0.61.2 ; python_version > "3.9" #Description: Just-In-Time Compiler for Numerical Functions #Pinned versions: 0.54.1, 0.49.0, <=0.49.1 #test that import: test_numba_integration.py @@ -135,6 +133,7 @@ numpy==2.0.2 ; python_version == "3.9" numpy==2.1.2 ; python_version > "3.9" pandas==2.2.3 + #onnxruntime #Description: scoring engine for Open Neural Network Exchange (ONNX) models #Pinned versions: 1.9.0 @@ -163,10 +162,11 @@ pillow==11.0.0 #Pinned versions: 10.3.0 #test that import: -protobuf==5.29.4 -#Description: Google's data interchange format -#Pinned versions: 5.29.4 -#test that import: test_tensorboard.py, test/onnx/* +protobuf==3.20.2 ; python_version <= "3.12" +protobuf==4.25.1 ; python_version == "3.13" +#Description: Google’s data interchange format +#Pinned versions: 3.20.1 +#test that import: test_tensorboard.py psutil #Description: information on running processes and system utilization @@ -263,11 +263,6 @@ tb-nightly==2.13.0a20230426 #Pinned versions: #test that import: -tlparse==0.3.30 -#Description: parse logs produced by torch.compile -#Pinned versions: -#test that import: dynamo/test_structured_trace.py - # needed by torchgen utils typing-extensions>=4.10.0 #Description: type hints for python @@ -309,7 +304,7 @@ pytest-cpp==2.3.0 #Pinned versions: 2.3.0 #test that import: -z3-solver==4.15.1.0 +z3-solver==4.12.6.0 #Description: The Z3 Theorem Prover Project #Pinned versions: #test that import: @@ -326,7 +321,8 @@ pywavelets==1.7.0 ; python_version >= "3.12" #Pinned versions: 1.4.1 #test that import: -lxml==5.3.0 +lxml==5.3.0 ; python_version <= "3.12" +lxml==6.0.0 ; python_version == "3.13" #Description: This is a requirement of unittest-xml-reporting # Python-3.9 binaries @@ -338,8 +334,9 @@ sympy==1.13.3 #Pinned versions: #test that import: -onnx==1.18.0 -#Description: Required by onnx tests, and mypy and test_public_bindings.py when checking torch.onnx._internal +onnx==1.16.1 ; python_version <= "3.12" +onnx==1.18.0 ; python_version == "3.13" +#Description: Required by mypy and test_public_bindings.py when checking torch.onnx._internal #Pinned versions: #test that import: From 45f8d8e47f2355e2092954447e24221de11cbbf9 Mon Sep 17 00:00:00 2001 From: Prachi Gupta Date: Fri, 8 Aug 2025 21:16:15 +0000 Subject: [PATCH 6/8] Use newer versions from upstream --- .ci/docker/requirements-ci.txt | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/.ci/docker/requirements-ci.txt b/.ci/docker/requirements-ci.txt index 5cdcaa3d81a33..5d32a375ca7d7 100644 --- a/.ci/docker/requirements-ci.txt +++ b/.ci/docker/requirements-ci.txt @@ -168,6 +168,7 @@ protobuf==4.25.1 ; python_version == "3.13" #Pinned versions: 3.20.1 #test that import: test_tensorboard.py + psutil #Description: information on running processes and system utilization #Pinned versions: @@ -304,7 +305,7 @@ pytest-cpp==2.3.0 #Pinned versions: 2.3.0 #test that import: -z3-solver==4.12.6.0 +z3-solver==4.15.1.0 #Description: The Z3 Theorem Prover Project #Pinned versions: #test that import: @@ -334,9 +335,9 @@ sympy==1.13.3 #Pinned versions: #test that import: -onnx==1.16.1 ; python_version <= "3.12" -onnx==1.18.0 ; python_version == "3.13" -#Description: Required by mypy and test_public_bindings.py when checking torch.onnx._internal +onnx==1.18.0 +#Description: Required by onnx tests, and mypy and test_public_bindings.py when checking torch.onnx._internal + #Pinned versions: #test that import: From 44451cdaf470d4de5e4e38fd5b3afcec9350565c Mon Sep 17 00:00:00 2001 From: Prachi Gupta Date: Fri, 8 Aug 2025 21:17:44 +0000 Subject: [PATCH 7/8] Fix protobuf version --- .ci/docker/requirements-ci.txt | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/.ci/docker/requirements-ci.txt b/.ci/docker/requirements-ci.txt index 5d32a375ca7d7..9a7747095710a 100644 --- a/.ci/docker/requirements-ci.txt +++ b/.ci/docker/requirements-ci.txt @@ -162,10 +162,9 @@ pillow==11.0.0 #Pinned versions: 10.3.0 #test that import: -protobuf==3.20.2 ; python_version <= "3.12" -protobuf==4.25.1 ; python_version == "3.13" +protobuf==5.29.4 #Description: Google’s data interchange format -#Pinned versions: 3.20.1 +#Pinned versions: 5.29.4 #test that import: test_tensorboard.py From 1ee30ac39628b611d22a8fdf331e07b4a697ec5a Mon Sep 17 00:00:00 2001 From: Jithun Nair <37884920+jithunnair-amd@users.noreply.github.com> Date: Fri, 8 Aug 2025 16:20:29 -0500 Subject: [PATCH 8/8] typo --- .ci/docker/requirements-ci.txt | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/.ci/docker/requirements-ci.txt b/.ci/docker/requirements-ci.txt index 9a7747095710a..f12be92975054 100644 --- a/.ci/docker/requirements-ci.txt +++ b/.ci/docker/requirements-ci.txt @@ -163,9 +163,9 @@ pillow==11.0.0 #test that import: protobuf==5.29.4 -#Description: Google’s data interchange format +#Description: Google's data interchange format #Pinned versions: 5.29.4 -#test that import: test_tensorboard.py +#test that import: test_tensorboard.py, test/onnx/* psutil