[CI] remove verbose log flag for build

.github/scripts/generate_release_matrix.py

@@ -35,7 +35,7 @@ def as_matrix_item(self) -> dict[str, int | str]:
         return {
             "cuda": self.cuda,
             "torch": self.torch,
-            "python": self.python,
+            "python": format_python_version(self.python),
         }
 
 
@@ -57,6 +57,15 @@ def python_key(python_version: str) -> tuple[int, int]:
     return base, 0 if free_threaded else 1
 
 
+def format_python_version(python_version: str) -> str:
+    free_threaded = python_version.endswith("t")
+    base = python_version[:-1] if free_threaded else python_version
+    if len(base) < 2:
+        raise ValueError(f"unsupported python version format: {python_version}")
+    formatted = f"{base[0]}.{base[1:]}"
+    return f"{formatted}t" if free_threaded else formatted
+
+
 def fetch_torch_index(source_url: str) -> str:
     request = urllib.request.Request(
         source_url,

.github/scripts/run_tests.py

@@ -66,6 +66,7 @@ def worker() -> None:
 
 def stream_process_output(proc: subprocess.Popen[str], log_file: Path) -> int:
     assert proc.stdout is not None
+    log_file.parent.mkdir(parents=True, exist_ok=True)
     with log_file.open("w", encoding="utf-8") as fh:
         for line in proc.stdout:
             print(line, end="")

.github/workflows/release.yml

@@ -169,13 +169,15 @@ jobs:
         run: |
           echo "UV_PYTHON=${{ matrix.python }}" >> "$GITHUB_ENV"
           echo "UV_TORCH_BACKEND=cu${{ matrix.cuda }}" >> "$GITHUB_ENV"
+          export UV_PYTHON=${{ matrix.python }}
+          export UV_TORCH_BACKEND=cu${{ matrix.cuda }}
 
           env_name="cu${{ matrix.cuda }}_torch${{ matrix.torch }}_py${{ matrix.python }}_release"
-          /opt/uv/setup_uv_venv.sh $env_name ${{ matrix.python }}
+          source /opt/uv/setup_uv_venv.sh $env_name ${{ matrix.python }}
           python -VV
 
-          echo "source /opt/env/init_compiler_torch_only.sh ${{ matrix.cuda }} ${{ matrix.torch }} $UV_PYTHON"
-          source /opt/env/init_compiler_torch_only.sh ${{ matrix.cuda }} ${{ matrix.torch }} $UV_PYTHON
+          echo "source /opt/env/init_compiler_torch_only.sh ${{ matrix.cuda }} ${{ matrix.torch }} ${{ matrix.python }}"
+          /opt/env/init_compiler_torch_only.sh ${{ matrix.cuda }} ${{ matrix.torch }} ${{ matrix.python }}
 
       - name: Print uv env
         if: env.SHOULD_RUN == 1
@@ -232,7 +234,7 @@ jobs:
 
           [ $(stat -c%s "dist/$whl") -lt 104857600 ] && echo "wheel size < 100M" && exit 1 || echo "$whl size check passed."
 
-          if [ "${{ matrix.python }}" != "313t" ]; then # twine doesn't support python 3.13 yet
+          if [ "${{ matrix.python }}" != "3.13t" ]; then # twine doesn't support python 3.13 yet
             uv pip install twine
             twine check dist/$whl
           fi

gptqmodel/utils/awq.py

@@ -10,7 +10,6 @@
 
 from .cpp import TorchOpsJitExtension, default_torch_ops_build_root
 
-
 _AWQ_OPS_NAME = "gptqmodel_awq_ops"
 _AWQ_OPS_NAMESPACE = "gptqmodel_awq"
 
@@ -62,7 +61,7 @@ def _awq_extra_cuda_cflags() -> list[str]:
         os.getenv("NVCC_THREADS", "2"),
         "--optimize=3",
         "-Xptxas",
-        "-v,-O3,-dlcm=ca",
+        "-O3,-dlcm=ca",  # -v, removed, or log file is too large.
         "-lineinfo",
         "-Xfatbin",
         "-compress-all",

setup.py

@@ -15,7 +15,6 @@
 from setuptools import find_namespace_packages, find_packages, setup
 from setuptools.command.bdist_wheel import bdist_wheel as _bdist_wheel
 
-
 if Version(setuptools.__version__) < Version("78.1.1"):
     raise RuntimeError(
         f"\033[31mYour setuptools version (`{setuptools.__version__}`) is too old and incompatible.\n"
@@ -715,7 +714,7 @@ def _env_enabled_any(names, default="1") -> bool:
                 # Print register/shared-memory usage per kernel (debug aid, no perf effect)
                 # Ensure PTXAS uses maximum optimization
                 # Cache global loads in both L1 and L2 (better for memory-bound kernels)
-                "-Xptxas", "-v,-O3,-dlcm=ca",
+                "-Xptxas", "-O3,-dlcm=ca",  # -v, removed, or log file is too large.
                 "-lineinfo",  # keep source line info for profiling
                 # "--resource-usage",  # show per-kernel register/SMEM usage
                 "-Xfatbin", "-compress-all",  # compress fatbin