commit eb041346267f6710776a56e4ee49c5b682013dd1 · pyrox.dev/nixpkgs

-18

pkgs/development/python-modules/vllm/0004-drop-lsmod.patch

···

       1
       1
       -
       --- a/setup.py

     

       2
       2
       -
       +++ b/setup.py

     

       3
       3
       -
       @@ -340,14 +340,7 @@ def _is_hpu() -> bool:

     

       4
       4
       -
                out = subprocess.run(["hl-smi"], capture_output=True, check=True)

     

       5
       5
       -
                is_hpu_available = out.returncode == 0

     

       6
       6
       -
            except (FileNotFoundError, PermissionError, subprocess.CalledProcessError):

     

       7
       7
       -
       -        if sys.platform.startswith("linux"):

     

       8
       8
       -
       -            try:

     

       9
       9
       -
       -                output = subprocess.check_output(

     

       10
       10
       -
       -                    'lsmod | grep habanalabs | wc -l', shell=True)

     

       11
       11
       -
       -                is_hpu_available = int(output) > 0

     

       12
       12
       -
       -            except (ValueError, FileNotFoundError, PermissionError,

     

       13
       13
       -
       -                    subprocess.CalledProcessError):

     

       14
       14
       -
       -                pass

     

       15
       15
       -
       +        is_hpu_available = False

     

       16
       16
       -
            return is_hpu_available

     

       17
       17
       -
        

     

       18
       18
       -

+6 -19

pkgs/development/python-modules/vllm/0005-drop-intel-reqs.patch

···

       1
       1
       -
       From 7511784ceb9252091a9d63ac6b54dcc67dd2b262 Mon Sep 17 00:00:00 2001

     

       2
       2
       -
       From: Conroy Cheers <conroy@corncheese.org>

     

       3
       3
       -
       Date: Fri, 13 Jun 2025 17:42:10 +1000

     

       4
       4
       -
       Subject: [PATCH] drop intel reqs

     

       5
       5
       -
       

     

       6
       6
       -
       ---

     

       7
       7
       -
        requirements/cpu.txt | 3 ---

     

       8
       8
       -
        1 file changed, 3 deletions(-)

     

       9
       9
       -
       

     

       10
       1
        
       diff --git a/requirements/cpu.txt b/requirements/cpu.txt

     

       11
       11
       -
       index d7b0fc6d8..be2df751b 100644

     

       2
       2
       +
       index d80354342..7434f32f0 100644

     

       12
       3
        
       --- a/requirements/cpu.txt

     

       13
       4
        
       +++ b/requirements/cpu.txt

     

       14
       14
       -
       @@ -24,8 +24,5 @@ datasets # for benchmark scripts

     

       15
       15
       -
        # cpu cannot use triton 3.3.0

     

       16
       16
       -
        triton==3.2.0; platform_machine == "x86_64"

     

       5
       5
       +
       @@ -21,7 +21,4 @@ torchvision; platform_machine != "ppc64le" and platform_machine != "s390x"

     

       6
       6
       +
        torchvision==0.22.0; platform_machine == "ppc64le"

     

       7
       7
       +
        datasets # for benchmark scripts

     

       17
       8
        
        

     

       18
       9
        
       -# Intel Extension for PyTorch, only for x86_64 CPUs

     

       19
       10
        
       -intel-openmp==2024.2.1; platform_machine == "x86_64"

     

       20
       20
       -
       -intel_extension_for_pytorch==2.7.0; platform_machine == "x86_64"

     

       21
       21
       -
        py-libnuma; platform_system != "Darwin"

     

       22
       22
       -
        psutil; platform_system != "Darwin"

     

       23
       23
       -
       -- 

     

       24
       24
       -
       2.49.0

     

       25
       25
       -
       

     

       11
       11
       +
       -intel_extension_for_pytorch==2.6.0; platform_machine == "x86_64" # torch>2.6.0+cpu has performance regression on x86 platform, see https://github.com/pytorch/pytorch/pull/151218

     

       12
       12
       +
        triton==3.2.0; platform_machine == "x86_64" # Triton is required for torch 2.6+cpu, as it is imported in torch.compile.

+19 -21

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

···

       5
       5
        
         buildPythonPackage,

     

       6
       6
        
         pythonAtLeast,

     

       7
       7
        
         fetchFromGitHub,

     

       8
       8
       -
         fetchpatch,

     

       9
       8
        
         symlinkJoin,

     

       10
       9
        
         autoAddDriverRunpath,

     

       10
       10
       +
         fetchpatch2,

     

       11
       11
        
       

     

       12
       12
        
         # build system

     

       13
       13
        
         cmake,

     
···

       45
       45
        
         lm-format-enforcer,

     

       46
       46
        
         prometheus-fastapi-instrumentator,

     

       47
       47
        
         cupy,

     

       48
       48
       +
         cbor2,

     

       49
       49
       +
         pybase64,

     

       48
       50
        
         gguf,

     

       49
       51
        
         einops,

     

       50
       52
        
         importlib-metadata,

     
···

       97
       99
        
         cutlass = fetchFromGitHub {

     

       98
       100
        
           owner = "NVIDIA";

     

       99
       101
        
           repo = "cutlass";

     

       100
       100
       -
           tag = "v3.9.2";

     

       101
       101
       -
           hash = "sha256-teziPNA9csYvhkG5t2ht8W8x5+1YGGbHm8VKx4JoxgI=";

     

       102
       102
       +
           tag = "v4.0.0";

     

       103
       103
       +
           hash = "sha256-HJY+Go1viPkSVZPEs/NyMtYJzas4mMLiIZF3kNX+WgA=";

     

       102
       104
        
         };

     

       103
       105
        
       

     

       104
       106
        
         flashmla = stdenv.mkDerivation {

     
···

       138
       140
        
           src = fetchFromGitHub {

     

       139
       141
        
             owner = "vllm-project";

     

       140
       142
        
             repo = "flash-attention";

     

       141
       141
       -
             rev = "8798f27777fb57f447070301bf33a9f9c607f491";

     

       142
       142
       -
             hash = "sha256-UTUvATGN1NU/Bc8qo078q6bEgILLmlrjL7Yk2iAJhg4=";

     

       143
       143
       +
             rev = "1c2624e53c078854e0637ee566c72fe2107e75f4";

     

       144
       144
       +
             hash = "sha256-WWFhHEUSAlsXr2yR4rGlTQQnSafXKg8gO5PQA8HPYGE=";

     

       143
       145
        
           };

     

       144
       146
        
       

     

       145
       147
        
           dontConfigure = true;

     
···

       161
       163
        
       

     

       162
       164
        
         cpuSupport = !cudaSupport && !rocmSupport;

     

       163
       165
        
       

     

       164
       164
       -
         # https://github.com/pytorch/pytorch/blob/v2.7.0/torch/utils/cpp_extension.py#L2343-L2345

     

       166
       166
       +
         # https://github.com/pytorch/pytorch/blob/v2.7.1/torch/utils/cpp_extension.py#L2343-L2345

     

       165
       167
        
         supportedTorchCudaCapabilities =

     

       166
       168
        
           let

     

       167
       169
        
             real = [

     
···

       247
       249
        
       

     

       248
       250
        
       buildPythonPackage rec {

     

       249
       251
        
         pname = "vllm";

     

       250
       250
       -
         version = "0.9.1";

     

       252
       252
       +
         version = "0.10.0";

     

       251
       253
        
         pyproject = true;

     

       252
       254
        
       

     

       253
       255
        
         # https://github.com/vllm-project/vllm/issues/12083

     
···

       259
       261
        
           owner = "vllm-project";

     

       260
       262
        
           repo = "vllm";

     

       261
       263
        
           tag = "v${version}";

     

       262
       262
       -
           hash = "sha256-sp7rDpewTPXTVRBJHJMj+8pJDS6wAu0/OTJZwbPPqKc=";

     

       264
       264
       +
           hash = "sha256-R9arpFz+wkDGmB3lW+H8d/37EoAQDyCWjLHJW1VTutk=";

     

       263
       265
        
         };

     

       264
       266
        
       

     

       265
       267
        
         patches = [

     

       266
       266
       -
           (fetchpatch {

     

       267
       267
       -
             name = "remove-unused-opentelemetry-semantic-conventions-ai-dep.patch";

     

       268
       268
       -
             url = "https://github.com/vllm-project/vllm/commit/6a5d7e45f52c3a13de43b8b4fa9033e3b342ebd2.patch";

     

       269
       269
       -
             hash = "sha256-KYthqu+6XwsYYd80PtfrMMjuRV9+ionccr7EbjE4jJE=";

     

       270
       270
       -
           })

     

       271
       271
       -
           (fetchpatch {

     

       272
       272
       -
             name = "fall-back-to-gloo-when-nccl-unavailable.patch";

     

       273
       273
       -
             url = "https://github.com/vllm-project/vllm/commit/aa131a94410683b0a02e74fed2ce95e6c2b6b030.patch";

     

       274
       274
       -
             hash = "sha256-jNlQZQ8xiW85JWyBjsPZ6FoRQsiG1J8bwzmQjnaWFBg=";

     

       268
       268
       +
           # error: ‘BF16Vec16’ in namespace ‘vec_op’ does not name a type; did you mean ‘FP16Vec16’?

     

       269
       269
       +
           # Reported: https://github.com/vllm-project/vllm/issues/21714

     

       270
       270
       +
           # Fix from https://github.com/vllm-project/vllm/pull/21848

     

       271
       271
       +
           (fetchpatch2 {

     

       272
       272
       +
             name = "build-fix-for-arm-without-bf16";

     

       273
       273
       +
             url = "https://github.com/vllm-project/vllm/commit/b876860c6214d03279e79e0babb7eb4e3e286cbd.patch";

     

       274
       274
       +
             hash = "sha256-tdBAObFxliVUNTWeSggaLtS4K9f8zEVu22nSgRmMsDs=";

     

       275
       275
        
           })

     

       276
       276
        
           ./0002-setup.py-nix-support-respect-cmakeFlags.patch

     

       277
       277
        
           ./0003-propagate-pythonpath.patch

     

       278
       278
       -
           ./0004-drop-lsmod.patch

     

       279
       278
        
           ./0005-drop-intel-reqs.patch

     

       280
       279
        
         ];

     

       281
       280
        
       

     
···

       354
       353
        
           aioprometheus

     

       355
       354
        
           blake3

     

       356
       355
        
           cachetools

     

       356
       356
       +
           cbor2

     

       357
       357
        
           depyf

     

       358
       358
        
           fastapi

     

       359
       359
        
           llguidance

     
···

       366
       366
        
           prometheus-fastapi-instrumentator

     

       367
       367
        
           py-cpuinfo

     

       368
       368
        
           pyarrow

     

       369
       369
       +
           pybase64

     

       369
       370
        
           pydantic

     

       370
       371
        
           python-json-logger

     

       371
       372
        
           python-multipart

     
···

       476
       477
        
             # find_isa

     

       477
       478
        
             "x86_64-darwin"

     

       478
       479
        
           ];

     

       479
       479
       -
           # ValueError: 'aimv2' is already used by a Transformers config, pick another name.

     

       480
       480
       -
           # Version bump ongoing in https://github.com/NixOS/nixpkgs/pull/429117

     

       481
       481
       -
           broken = true;

     

       482
       480
        
         };

     

       483
       481
        
       }