move req.txt into toml

.github/workflows/release.yml

@@ -53,7 +53,7 @@ on:
 
 env:
   CUDA_DEVICE_ORDER: PCI_BUS_ID
-  RUNNER: 10.0.14.248
+  RUNNER: 10.0.13.31
   TORCH_CUDA_ARCH_LIST: '8.0 8.6 8.9 9.0 12.0'
   CUDA_ARCH_LIST: '8.0 8.6 8.9 9.0 12.0'
   RELEASE_MODE: 1
@@ -63,7 +63,7 @@ env:
   ref: ${{ github.event.inputs.ref || github.ref }}
   MAX_JOBS: 4
   GPTQMODEL_BUILD_QQQ: 0
-  GTTQMODEL_BUILD_EORA: 0
+  GPTQMODEL_BUILD_EORA: 0
   GPTQMODEL_BUILD_EXLLAMA_V1: 0
 
 concurrency:
@@ -287,7 +287,7 @@ jobs:
           echo "== python =="
           python --version
           echo "== nvcc =="
-          #nvcc --version
+          nvcc --version
           echo "== torch =="
           pip show torch
 
@@ -336,6 +336,8 @@ jobs:
       - check-vm
     container:
       image: ${{ needs.check-vm.outputs.ip }}:5000/nvidia/cuda:128-ubuntu22.04_0919
+      volumes:
+        - /home/ci/models/pyenv:/opt/pyenv
     env:
       RELEASE_MODE: 0
     steps:
@@ -345,6 +347,21 @@ jobs:
           repository: ${{ env.repo }}
           ref: ${{ env.ref }}
 
+      - name: Print Env
+        run: |
+          env_name="py313_release_source"
+          if [ -d "$(pyenv root)/versions/$env_name" ]; then
+            echo "env: ${env_name} exists, skip"
+            pyenv local $env_name
+            pyenv activate $env_name
+          else
+            echo "creating venv: ${env_name}..."
+            pyenv virtualenv 3.13 "$env_name" || true
+            pyenv local $env_name
+            pyenv activate $env_name
+            bash -c "$(curl -L http://${RUNNER}/scripts/env/init_compiler.sh)" @ 128 2.8.0 313
+          fi
+
       - name: Fetch PR by number
         if: ${{ github.event.inputs.pr_number != 0 }}
         run: |
@@ -353,6 +370,7 @@ jobs:
           git config --global --add safe.directory $(pwd)
           git fetch origin pull/${PR_NUMBER}/head:pr-${PR_NUMBER}
           git checkout pr-${PR_NUMBER}
+
       - name: Install requirements
         run: pip install build setuptools -U -i http://${{ needs.check-vm.outputs.ip }}/simple/ --trusted-host ${{ needs.check-vm.outputs.ip }}
 

.github/workflows/unit_tests.yml

@@ -59,12 +59,13 @@ env:
   CUDA_VISIBLE_DEVICES: 0
   TORCH_CUDA_ARCH_LIST: '8.6 8.9 9.0 12.0'
   PYTORCH_CUDA_ALLOC_CONF: 'expandable_segments:True'
-  MAX_JOBS: 8
+  MAX_JOBS: 4
   RUNNER: 10.0.13.31
   XEON5: 10.0.14.248
   CUDA_VERSION: 128
   TORCH_VERSION: 2.8.0
-  PYTHON_VERSION: 312
+  PYTHON_VERSION: 313
+  # PYTHON_GIL: 0 // test libs don't support yet
   LEGACY_TESTS: "models/test_internlm.py,models/test_internlm2_5.py,models/test_xverse.py"
   IGNORED_TEST_FILES: "test_tgi.py,test_gptneox.py,models/test_mixtral.py,models/test_phi_3_moe.py,test_bits_new.py"
   GPTQMODEL_FORCE_BUILD: 1
@@ -521,9 +522,6 @@ jobs:
             uv pip install numpy==1.26.3
           fi
 
-          echo "===== install -r requirements.txt ====="
-          uv pip install -r requirements.txt
-
           echo "===== uninstall torchao ====="
           uv pip uninstall torchao
 
@@ -643,9 +641,6 @@ jobs:
           echo "=== installing test tools"
           uv pip install pytest parameterized vllm lm-eval device-smi mlx-lm -U -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
 
-          echo "=== installing requirements"
-          uv pip install -r requirements.txt -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
-
           echo "=== installing gptqmodel"
           uv pip install . --no-build-isolation -i http://$RUNNER/simple/ --trusted-host $RUNNER --extra-index-url https://pypi.org/simple
 

MANIFEST.in

@@ -3,6 +3,5 @@ global-include gptqmodel_ext/**/*.h
 global-include gptqmodel_ext/**/*.cpp
 global-include gptqmodel_ext/**/*.cu
 global-include gptqmodel_ext/**/*.py
-include requirements.txt
 prune tests/
 prune format/

pyproject.toml

@@ -1,5 +1,7 @@
 [build-system]
-requires = ["setuptools >= 64"]
+requires = [
+    "setuptools >= 64",
+]
 build-backend = "setuptools.build_meta:__legacy__"
 
 [project]
@@ -25,10 +27,28 @@ classifiers = [
     "Topic :: Scientific/Engineering :: Artificial Intelligence",
     "Topic :: Scientific/Engineering :: Information Analysis",
 ]
-dynamic = ["version", "dependencies"]
-
-[tool.setuptools.dynamic]
-dependencies = {file = ["requirements.txt"]}
+dynamic = ["version"]
+dependencies = [
+    "accelerate>=1.10.1",
+    # "datasets>=3.5.0",
+    "numpy>=2.2.6",
+    "torch>=2.8.0",
+    "safetensors>=0.6.2",
+    "transformers>=4.56.0",
+    "threadpoolctl>=3.6.0",
+    "packaging>=24.2",
+    "device-smi==0.4.1",
+    "protobuf>=6.32.0",
+    "pillow>=11.3.0",
+    "hf_transfer>=0.1.9",
+    "huggingface_hub>=0.34.4",
+    "random_word==1.0.13",
+    "tokenicer>=0.0.5",
+    "logbar==0.0.4",
+    # "soundfile==0.13.1",   # Qwen-Omni 可选依赖
+    "wheel>=0.45.1",
+    "maturin>=1.9.3",
+]
 
 [project.urls]
 Homepage = "https://github.com/ModelCloud/GPTQModel"

setup.py

@@ -184,8 +184,10 @@ def _detect_torch_version() -> str | None:
 
 
 def _major_minor(v: str) -> str:
-    parts = v.split(".")
-    return ".".join(parts[:2]) if parts else v
+    if v:
+        parts = v.split(".")
+        return ".".join(parts[:2]) if parts else v
+    return v
 
 
 def _detect_cuda_version() -> str | None:
@@ -333,11 +335,6 @@ def _resolve_wheel_url(tag_name: str, wheel_name: str) -> str:
     # Fallback: default GitHub template
     return DEFAULT_WHEEL_URL_TEMPLATE.format(tag_name=tag_name, wheel_name=wheel_name)
 
-requirements = []
-if not os.getenv("CI"):
-    with open("requirements.txt", encoding="utf-8") as f:
-        requirements = [line.strip() for line in f if line.strip()]
-
 # Decide HAS_CUDA_V8 without torch
 HAS_CUDA_V8 = False
 if CUDA_ARCH_LIST: