python312Packages.torch: 2.6.0 -> 2.7.0 (#401807)

pkgs/development/python-modules/torch/source/default.nix

···
   # Build inputs
   apple-sdk_13,
   numactl,
+  llvmPackages,
 
   # dependencies
   astunparse,
···
 
   setBool = v: if v then "1" else "0";
 
-  # https://github.com/pytorch/pytorch/blob/v2.6.0/torch/utils/cpp_extension.py#L2046-L2048
+  # https://github.com/pytorch/pytorch/blob/v2.7.0/torch/utils/cpp_extension.py#L2343-L2345
   supportedTorchCudaCapabilities =
     let
       real = [
···
         "9.0"
         "9.0a"
         "10.0"
+        "10.0"
+        "10.0a"
+        "10.1"
+        "10.1a"
+        "12.0"
+        "12.0a"
       ];
       ptx = lists.map (x: "${x}+PTX") real;
     in
···
     else if cudaSupport then
       gpuArchWarner supportedCudaCapabilities unsupportedCudaCapabilities
     else if rocmSupport then
-      rocmPackages.clr.gpuTargets
+      # Remove RDNA1 gfx101x archs from default ROCm support list to avoid
+      # use of undeclared identifier 'CK_BUFFER_RESOURCE_3RD_DWORD'
+      # TODO: Retest after ROCm 6.4 or torch 2.8
+      lib.lists.subtractLists [
+        "gfx1010"
+        "gfx1012"
+      ] (rocmPackages.clr.localGpuTargets or rocmPackages.clr.gpuTargets)
     else
       throw "No GPU targets specified"
   );
···
       rccl
       miopen
       aotriton
+      composable_kernel
       rocrand
       rocblas
       rocsparse
···
 buildPythonPackage rec {
   pname = "torch";
   # Don't forget to update torch-bin to the same version.
-  version = "2.6.0";
+  version = "2.7.0";
   pyproject = true;
 
   stdenv = stdenv';
···
   patches =
     [
       ./clang19-template-warning.patch
-      # fix invalid static cast in XNNPACK
-      # https://github.com/google/XNNPACK/issues/7489
-      ./xnnpack-bfloat16.patch
     ]
     ++ lib.optionals cudaSupport [ ./fix-cmake-cuda-toolkit.patch ]
     ++ lib.optionals stdenv.hostPlatform.isLinux [
···
 
   postPatch =
     ''
+      # Prevent NCCL from being cloned during the configure phase
+      # TODO: remove when updating to the next release as it will not be needed anymore
+      substituteInPlace tools/build_pytorch_libs.py \
+        --replace-fail "  checkout_nccl()" "  "
+
       substituteInPlace cmake/public/cuda.cmake \
         --replace-fail \
           'message(FATAL_ERROR "Found two conflicting CUDA' \
···
 
       # Replace hard-coded rocm paths
       substituteInPlace caffe2/CMakeLists.txt \
-        --replace-fail "/opt/rocm" "${rocmtoolkit_joined}" \
         --replace-fail "hcc/include" "hip/include" \
         --replace-fail "rocblas/include" "include/rocblas" \
         --replace-fail "hipsparse/include" "include/hipsparse"
 
       # Doesn't pick up the environment variable?
       substituteInPlace third_party/kineto/libkineto/CMakeLists.txt \
-        --replace-fail "\''$ENV{ROCM_SOURCE_DIR}" "${rocmtoolkit_joined}" \
-        --replace-fail "/opt/rocm" "${rocmtoolkit_joined}"
+        --replace-fail "\''$ENV{ROCM_SOURCE_DIR}" "${rocmtoolkit_joined}"
 
       # Strangely, this is never set in cmake
       substituteInPlace cmake/public/LoadHIP.cmake \
···
       blas
       blas.provider
     ]
+    ++ lib.optionals stdenv.cc.isClang [ llvmPackages.openmp ]
     ++ lib.optionals cudaSupport (
       with cudaPackages;
       [
···
         cusparselt
         libcublas
         libcufft
+        libcufile
         libcurand
         libcusolver
         libcusparse
···
         --replace-fail "\''${_IMPORT_PREFIX}/lib64" "$lib/lib"
 
       substituteInPlace $dev/share/cmake/ATen/ATenConfig.cmake \
-        --replace-fail "/build/source/torch/include" "$dev/include"
+        --replace-fail "/build/${src.name}/torch/include" "$dev/include"
     '';
 
   postFixup =
···
     homepage = "https://pytorch.org/";
     license = lib.licenses.bsd3;
     maintainers = with lib.maintainers; [
+      GaetanLepage
       teh
       thoughtpolice
       tscholak

pkgs/development/python-modules/torch/source/src.nix

···
   fetchFromGitHub,
   runCommand,
 }:
-assert version == "2.6.0";
+assert version == "2.7.0";
 (rec {
   src_asmjit = fetchFromGitHub {
     owner = "asmjit";
···
   src_composable_kernel = fetchFromGitHub {
     owner = "ROCm";
     repo = "composable_kernel";
-    rev = "50ee4267e27b875d149e642f4cebd47be1dc3b57";
-    hash = "sha256-COkyf+FZzX6OdOPCHkP2bXsVvSXK9UR9s7RHWRtIXUE=";
+    rev = "8086bbe3a78d931eb96fe12fdc014082e18d18d3";
+    hash = "sha256-fyL1SzRs5CXW5CWy6kCN1y1xX6cG+ur7iQlbKX2zbCM=";
+  };
+  src_composable_kernel_flash-attention = fetchFromGitHub {
+    owner = "ROCm";
+    repo = "composable_kernel";
+    rev = "888317e698e9803c62bd38568abc9e05d7709f33";
+    hash = "sha256-0FAiGf7AErBzGxhOFcLQ4ceshqp0Esqo4ee7NsjfJUo=";
   };
   src_cpp-httplib = fetchFromGitHub {
     owner = "yhirose";
···
   src_cudnn-frontend = fetchFromGitHub {
     owner = "NVIDIA";
     repo = "cudnn-frontend";
-    rev = "936021bfed8c91dc416af1588b2c4eca631a9e45";
-    hash = "sha256-hKqIWGxVco1qkKxDZjc+pUisIcYJwFjZobJZg1WgDvY=";
+    rev = "91b7532f3386768bba4f444ee7672b497f34da8a";
+    hash = "sha256-Ks07ApADA3xQQ+N5BIfvDG+djCvxYL9btF8Aw7TuMvQ=";
   };
   src_cutlass = fetchFromGitHub {
     owner = "NVIDIA";
     repo = "cutlass";
-    rev = "bbe579a9e3beb6ea6626d9227ec32d0dae119a49";
-    hash = "sha256-81O80F3MMOn22N9UaXLU6/9DTVWenYvKhLTHoxw8EEU=";
+    rev = "afa1772203677c5118fcd82537a9c8fefbcc7008";
+    hash = "sha256-oIzlbKRdOh6gp6nRZ8udLSqleBFoFtgM7liCBlHZLOk=";
   };
   src_cutlass_fbgemm = fetchFromGitHub {
     owner = "NVIDIA";
···
     rev = "fc9ebc645b63f3a6bc80aaefde5c063fb72110d6";
     hash = "sha256-e2SwXNNwjl/1fV64b+mOJvwGDYeO1LFcqZGbNten37U=";
   };
+  src_cutlass_flash-attention = fetchFromGitHub {
+    owner = "NVIDIA";
+    repo = "cutlass";
+    rev = "c506e16788cb08416a4a57e11a9067beeee29420";
+    hash = "sha256-P/BgJ7MdNTzNgsqoTjxrOQwvKfhAS+rmBw3sC/SdO1I=";
+  };
   src_DCGM = fetchFromGitHub {
     owner = "NVIDIA";
     repo = "DCGM";
···
     rev = "7e1e1fe3858c63c251c637ae41a20de425dde96f";
     hash = "sha256-PsgUHtCE3dNR2QdUnRjrXb0ZKZNGwFkA8RWYkZEklEY=";
   };
+  src_flash-attention = fetchFromGitHub {
+    owner = "Dao-AILab";
+    repo = "flash-attention";
+    rev = "979702c87a8713a8e0a5e9fee122b90d2ef13be5";
+    hash = "sha256-7jk8TD+EkkOPrVT5Nzoc+zoZ4jH/C1AtjLcP12hBrp0=";
+  };
   src_flatbuffers = fetchFromGitHub {
     owner = "google";
     repo = "flatbuffers";
···
   src_fmt = fetchFromGitHub {
     owner = "fmtlib";
     repo = "fmt";
-    rev = "0c9fce2ffefecfdce794e1859584e25877b7b592";
-    hash = "sha256-IKNt4xUoVi750zBti5iJJcCk3zivTt7nU12RIf8pM+0=";
+    rev = "123913715afeb8a437e6388b4473fcc4753e1c9a";
+    hash = "sha256-sUbxlYi/Aupaox3JjWFqXIjcaQa0LFjclQAOleT+FRA=";
   };
   src_fmt_dynolog = fetchFromGitHub {
     owner = "fmtlib";
···
   src_ideep = fetchFromGitHub {
     owner = "intel";
     repo = "ideep";
-    rev = "c7ccd5bdbe5434ba156f4e856dcef0601637334b";
-    hash = "sha256-/u3wJQl49tBw/QNVDBXL+eNSRCuvltso7g1+siIhlXM=";
+    rev = "719d8e6cd7f7a0e01b155657526d693acf97c2b3";
+    hash = "sha256-WX1lluqYeoB2IuwJX1vqZ6nr8W8KsWxUDIvKNA3CAxo=";
   };
   src_ittapi = fetchFromGitHub {
     owner = "intel";
···
   src_kineto = fetchFromGitHub {
     owner = "pytorch";
     repo = "kineto";
-    rev = "338140f58a28d599da3434ced4fd2d75dd1a213d";
-    hash = "sha256-ydHUYB2m7dbbqBtLordZUCraewU0Q4m/ohBXNKtqHnw=";
+    rev = "a054a4be0db117c579a21747debf19c863631f26";
+    hash = "sha256-FyxMiDoDrjI8JZZNxK4S18yjULkKLUvqLQsCyxaTLSg=";
+  };
+  src_kleidiai = fetchFromGitHub {
+    owner = "ARM-software";
+    repo = "kleidiai";
+    rev = "ef685a13cfbe8d418aa2ed34350e21e4938358b6";
+    hash = "sha256-5RS2o+163/6Q+0y0FuWz6OBXw3FKxZINDkGOqlmd2N0=";
   };
   src_libnop = fetchFromGitHub {
     owner = "google";
···
   src_mkl-dnn = fetchFromGitHub {
     owner = "intel";
     repo = "mkl-dnn";
-    rev = "66f0cb9eb66affd2da3bf5f8d897376f04aae6af";
-    hash = "sha256-/ERkk6bgGEKoJEVdnBxMFEzB8pii71t3zQZNtyg+TdQ=";
-  };
-  src_nccl = fetchFromGitHub {
-    owner = "NVIDIA";
-    repo = "nccl";
-    rev = "ab2b89c4c339bd7f816fbc114a4b05d386b66290";
-    hash = "sha256-IF2tILwW8XnzSmfn7N1CO7jXL95gUp02guIW5n1eaig=";
+    rev = "8d263e693366ef8db40acc569cc7d8edf644556d";
+    hash = "sha256-+4z5l0mJsw0SOW245GfZh41mdHGZ8u+xED7afm6pQjs=";
   };
   src_NNPACK = fetchFromGitHub {
     owner = "Maratyszcza";
···
     hash = "sha256-lV+VZi2b4SQlRYrhKx9Dxc6HlDEFz3newvcBjTekupo=";
   };
   src_pthreadpool = fetchFromGitHub {
-    owner = "google";
+    owner = "Maratyszcza";
     repo = "pthreadpool";
-    rev = "b92447772365661680f486e39a91dfe6675adafc";
-    hash = "sha256-lRY+5sYv/KOuVngsrODRa4/yj1CDmZBW2HuKEbjsY+0=";
+    rev = "4fe0e1e183925bf8cfa6aae24237e724a96479b8";
+    hash = "sha256-R4YmNzWEELSkAws/ejmNVxqXDTJwcqjLU/o/HvgRn2E=";
   };
   src_pybind11 = fetchFromGitHub {
     owner = "pybind";
···
   src_pytorch = fetchFromGitHub {
     owner = "pytorch";
     repo = "pytorch";
-    rev = "v2.6.0";
-    hash = "sha256-X58OPtnPplEFtzOrQzyYRwdTmaoWEOGWWF5v6y+vaWo=";
+    rev = "v2.7.0";
+    hash = "sha256-ReXyzy+OuYxEQwU+t2WL3+jqd7ItdW6w8MiS0f9t+aY=";
   };
   src_sleef = fetchFromGitHub {
     owner = "shibatch";
     repo = "sleef";
-    rev = "60e76d2bce17d278b439d9da17177c8f957a9e9b";
-    hash = "sha256-JfARLkdt4je8ll+oqPGJqzUCQbsXoJ0bbX3jf0aHd0o=";
+    rev = "56e1f79cb140fb9326d612d0be06b5250565cade";
+    hash = "sha256-5hha7c/Lu6fkbXGlDieoJP6n2bnjY5iPp+hm0f//0ek=";
   };
   src_tensorpipe = fetchFromGitHub {
     owner = "pytorch";
···
   src_XNNPACK = fetchFromGitHub {
     owner = "google";
     repo = "XNNPACK";
-    rev = "4ea82e595b36106653175dcb04b2aa532660d0d8";
-    hash = "sha256-5IubxhCW5E6xzrOLfJzWPrPeftaxXApTx09p6B3NPgo=";
+    rev = "51a0103656eff6fc9bfd39a4597923c4b542c883";
+    hash = "sha256-nhowllqv/hBs7xHdTwbWtiKJ1mvAYsVIyIZ35ZGsmkg=";
   };
   src_asmjit_recursive = src_asmjit;
   src_benchmark_recursive = src_benchmark;
···
   src_civetweb_recursive = src_civetweb;
   src_clang-cindex-python3_recursive = src_clang-cindex-python3;
   src_composable_kernel_recursive = src_composable_kernel;
+  src_composable_kernel_flash-attention_recursive = src_composable_kernel_flash-attention;
   src_cpp-httplib_recursive = src_cpp-httplib;
   src_cpr_recursive = src_cpr;
   src_cpuinfo_recursive = src_cpuinfo;
···
   src_cudnn-frontend_recursive = src_cudnn-frontend;
   src_cutlass_recursive = src_cutlass;
   src_cutlass_fbgemm_recursive = src_cutlass_fbgemm;
+  src_cutlass_flash-attention_recursive = src_cutlass_flash-attention;
   src_DCGM_recursive = src_DCGM;
   src_dynolog_recursive = runCommand "dynolog" { } ''
     cp -r ${src_dynolog} $out
···
     cp -r ${src_hipify_torch_recursive}/* $out/third_party/hipify_torch
   '';
   src_fbjni_recursive = src_fbjni;
+  src_flash-attention_recursive = runCommand "flash-attention" { } ''
+    cp -r ${src_flash-attention} $out
+    chmod u+w $out/csrc/composable_kernel
+    cp -r ${src_composable_kernel_flash-attention_recursive}/* $out/csrc/composable_kernel
+    chmod u+w $out/csrc/cutlass
+    cp -r ${src_cutlass_flash-attention_recursive}/* $out/csrc/cutlass
+  '';
   src_flatbuffers_recursive = src_flatbuffers;
   src_fmt_recursive = src_fmt;
   src_fmt_dynolog_recursive = src_fmt_dynolog;
···
     chmod u+w $out/libkineto/third_party/googletest
     cp -r ${src_googletest_kineto_recursive}/* $out/libkineto/third_party/googletest
   '';
+  src_kleidiai_recursive = src_kleidiai;
   src_libnop_recursive = src_libnop;
   src_libuv_recursive = src_libuv;
   src_mimalloc_recursive = src_mimalloc;
   src_mkl-dnn_recursive = src_mkl-dnn;
-  src_nccl_recursive = src_nccl;
   src_NNPACK_recursive = src_NNPACK;
   src_NVTX_recursive = src_NVTX;
   src_onnx_recursive = runCommand "onnx" { } ''
···
     cp -r ${src_eigen_recursive}/* $out/third_party/eigen
     chmod u+w $out/third_party/fbgemm
     cp -r ${src_fbgemm_recursive}/* $out/third_party/fbgemm
+    chmod u+w $out/third_party/flash-attention
+    cp -r ${src_flash-attention_recursive}/* $out/third_party/flash-attention
     chmod u+w $out/third_party/flatbuffers
     cp -r ${src_flatbuffers_recursive}/* $out/third_party/flatbuffers
     chmod u+w $out/third_party/fmt
···
     cp -r ${src_ittapi_recursive}/* $out/third_party/ittapi
     chmod u+w $out/third_party/kineto
     cp -r ${src_kineto_recursive}/* $out/third_party/kineto
+    chmod u+w $out/third_party/kleidiai
+    cp -r ${src_kleidiai_recursive}/* $out/third_party/kleidiai
     chmod u+w $out/third_party/mimalloc
     cp -r ${src_mimalloc_recursive}/* $out/third_party/mimalloc
-    chmod u+w $out/third_party/nccl/nccl
-    cp -r ${src_nccl_recursive}/* $out/third_party/nccl/nccl
     chmod u+w $out/third_party/nlohmann
     cp -r ${src_json_recursive}/* $out/third_party/nlohmann
     chmod u+w $out/third_party/NNPACK

pkgs/development/python-modules/torch/source/xnnpack-bfloat16.patch

···
-diff --git a/third_party/XNNPACK/src/reference/unary-elementwise.cc b/third_party/XNNPACK/src/reference/unary-elementwise.cc
-index bd95ded6c..da892d8be 100644
---- a/third_party/XNNPACK/src/reference/unary-elementwise.cc
-+++ b/third_party/XNNPACK/src/reference/unary-elementwise.cc
-@@ -127,6 +127,16 @@ struct ConvertOp {
-   }
- };
- 
-+#ifdef XNN_HAVE_FLOAT16
-+template <>
-+struct ConvertOp<xnn_bfloat16, _Float16> {
-+  explicit ConvertOp(const xnn_unary_uparams*) {}
-+  _Float16 operator()(xnn_bfloat16 x) const {
-+    return static_cast<_Float16>(static_cast<float>(x));
-+  }
-+};
-+#endif
-+
- template <typename TIn, typename TOut>
- const xnn_unary_elementwise_config* get_convert_config(
-     std::true_type /*input_quantized*/, std::true_type /*output_quantized*/) {

pkgs/development/python-modules/torchaudio/default.nix

···
   pybind11,
   sox,
   torch,
+  llvmPackages,
 
   cudaSupport ? torch.cudaSupport,
   cudaPackages,
···
 in
 buildPythonPackage rec {
   pname = "torchaudio";
-  version = "2.6.0";
+  version = "2.7.0";
   pyproject = true;
 
   stdenv = torch.stdenv;
···
     owner = "pytorch";
     repo = "audio";
     tag = "v${version}";
-    hash = "sha256-WNdDBB2nShbPPW7GU5cMij00u5PUdN+j5pm41yrKnCA=";
+    hash = "sha256-/5XIVj0jLE7+A1LZxA3bFH3mdwNIcrV4XMOa4xznr/w=";
   };
 
   patches = [
···
     pybind11
     sox
     torch.cxxdev
-  ];
+  ] ++ lib.optionals stdenv.cc.isClang [ llvmPackages.openmp ];
 
   dependencies = [ torch ];
 
···
     platforms =
       lib.platforms.linux
       ++ lib.optionals (!cudaSupport && !rocmSupport) lib.platforms.darwin;
-    maintainers = with lib.maintainers; [ junjihashimoto ];
+    maintainers = with lib.maintainers; [
+      GaetanLepage
+      junjihashimoto
+    ];
   };
 }

pkgs/development/python-modules/torchvision/default.nix

···
   inherit (torch) cudaCapabilities cudaPackages cudaSupport;
 
   pname = "torchvision";
-  version = "0.21.0";
+  version = "0.22.0";
 in
 buildPythonPackage {
   inherit pname version;
···
     owner = "pytorch";
     repo = "vision";
     tag = "v${version}";
-    hash = "sha256-eDWw1Lt/sUc2Xt6cqOM5xaOfmsm+NEL5lZO+cIJKMtU=";
+    hash = "sha256-+70Rhfma4dM5tRlYNz0cuuTIxRbYf6dsnAhvkw7a5kM=";
   };
 
   nativeBuildInputs = [
···
     scipy
   ];
 
-  preConfigure =
-    ''
-      export TORCHVISION_INCLUDE="${libjpeg_turbo.dev}/include/"
-      export TORCHVISION_LIBRARY="${libjpeg_turbo}/lib/"
-    ''
-    + lib.optionalString cudaSupport ''
-      export TORCH_CUDA_ARCH_LIST="${lib.concatStringsSep ";" cudaCapabilities}"
-      export FORCE_CUDA=1
-    '';
+  env =
+    {
+      TORCHVISION_INCLUDE = "${libjpeg_turbo.dev}/include/";
+      TORCHVISION_LIBRARY = "${libjpeg_turbo}/lib/";
+    }
+    // lib.optionalAttrs cudaSupport {
+      TORCH_CUDA_ARCH_LIST = "${lib.concatStringsSep ";" cudaCapabilities}";
+      FORCE_CUDA = 1;
+    };
 
   # tests download big datasets, models, require internet connection, etc.
   doCheck = false;
···
     changelog = "https://github.com/pytorch/vision/releases/tag/v${version}";
     license = lib.licenses.bsd3;
     platforms = with lib.platforms; linux ++ lib.optionals (!cudaSupport) darwin;
-    maintainers = with lib.maintainers; [ ];
+    maintainers = with lib.maintainers; [ GaetanLepage ];
   };
 }

pkgs/development/rocm-modules/6/aotriton/default.nix

···
     gpuTargets' = lib.concatStringsSep ";" gpuTargets;
     compiler = "amdclang++";
     cFlags = "-O3 -DNDEBUG";
+    cxxFlags = "${cFlags} -Wno-c++11-narrowing";
     triton-llvm' = triton-llvm;
   in
   {
     pname = "aotriton";
-    version = "0.8.2b";
+    version = "0.9.2b";
 
     src = fetchFromGitHub {
       owner = "ROCm";
       repo = "aotriton";
       rev = "${finalAttrs.version}";
-      hash = "sha256-gSzGYWfyUNLyzqpu3BM8rjFFL7cRVZ+w9L5pnh9QGz4=";
+      hash = "sha256-1Cf0olD3zRg9JESD6s/WaGifm3kfD12VUvjTZHpmGAE=";
       fetchSubmodules = true;
     };
     env.CXX = compiler;
···
       substituteInPlace third_party/triton/python/setup.py \
         --replace-fail "from distutils.command.clean import clean" "import setuptools;from distutils.command.clean import clean" \
         --replace-fail 'system == "Linux"' 'False'
+      # Fix 'ld: error: unable to insert .comment after .comment'
+      substituteInPlace v2python/ld_script.py \
+        --replace-fail 'INSERT AFTER .comment;' ""
 
       cmakeFlagsArray+=(
         '-DCMAKE_C_FLAGS_RELEASE=${cFlags}'
-        '-DCMAKE_CXX_FLAGS_RELEASE=${cFlags}'
+        '-DCMAKE_CXX_FLAGS_RELEASE=${cxxFlags}'
       )
       prependToVar cmakeFlags "-GNinja"
       mkdir -p /build/tmp-home