Add support all the way from py3.10, bring back cublas dep

Author: Kosinkadink

.github/workflows/build-wheels.yml

@@ -35,8 +35,9 @@ jobs:
           # CUDA 13.0 will be installed inside the container
           CIBW_MANYLINUX_X86_64_IMAGE: quay.io/pypa/manylinux_2_28_x86_64
 
-          # Build only for Python 3.12 with stable ABI
-          CIBW_BUILD: "cp312-manylinux_x86_64"
+          # Build for Python 3.10, 3.11, and 3.12
+          # 3.10/3.11 will be version-specific, 3.12 will be abi3 (stable ABI)
+          CIBW_BUILD: "cp310-manylinux_x86_64 cp311-manylinux_x86_64 cp312-manylinux_x86_64"
           CIBW_SKIP: "*-musllinux_*"
 
           # Install CUDA 13.0 toolkit inside the container
@@ -109,8 +110,11 @@ jobs:
           if-no-files-found: error
 
   build_wheels_windows:
-    name: Build Windows wheel
+    name: Build Windows wheel (Python ${{ matrix.python-version }})
     runs-on: windows-2022
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
 
     steps:
       - name: Checkout code
@@ -219,12 +223,13 @@ jobs:
           
           Write-Host "CUDA MSBuild integration configured successfully"
 
-      - name: Set up Python 3.12
+      - name: Set up Python ${{ matrix.python-version }}
         uses: actions/setup-python@v5
         with:
-          python-version: "3.12"
+          python-version: ${{ matrix.python-version }}
 
       - name: Create python3.lib for stable ABI
+        if: matrix.python-version == '3.12'
         shell: pwsh
         run: |
           # Create python3.lib for stable ABI support
@@ -254,10 +259,11 @@ jobs:
           python -m pip install --upgrade pip
           pip install build setuptools wheel nanobind
 
-      - name: Build CUDA wheel (Stable ABI - cp312-abi3)
+      - name: Build CUDA wheel
         shell: bash
         run: |
           # Build with default platform-specific CUDA architectures from setup.py
+          # Python 3.12 will produce abi3 wheel, 3.10/3.11 will be version-specific
           python setup.py bdist_wheel
 
       - name: List built wheels
@@ -270,23 +276,26 @@ jobs:
       - name: Upload Windows wheel
         uses: actions/upload-artifact@v4
         with:
-          name: wheels-windows-cuda
+          name: wheels-windows-cuda-${{ matrix.python-version }}
           path: dist/*.whl
           if-no-files-found: error
 
   test_linux:
-    name: Test on Linux
+    name: Test on Linux (Python ${{ matrix.python-version }})
     needs: build_wheels_linux
     runs-on: ubuntu-22.04
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
 
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
 
-      - name: Set up Python 3.12
+      - name: Set up Python ${{ matrix.python-version }}
         uses: actions/setup-python@v5
         with:
-          python-version: "3.12"
+          python-version: ${{ matrix.python-version }}
 
       - name: Download CPU-only wheel
         uses: actions/download-artifact@v4
@@ -317,23 +326,26 @@ jobs:
           python -m pytest tests/ -v --tb=short
 
   test_windows:
-    name: Test on Windows
+    name: Test on Windows (Python ${{ matrix.python-version }})
     needs: build_wheels_windows
     runs-on: windows-2022
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
 
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
 
-      - name: Set up Python 3.12
+      - name: Set up Python ${{ matrix.python-version }}
         uses: actions/setup-python@v5
         with:
-          python-version: "3.12"
+          python-version: ${{ matrix.python-version }}
 
       - name: Download Windows wheel
         uses: actions/download-artifact@v4
         with:
-          name: wheels-windows-cuda
+          name: wheels-windows-cuda-${{ matrix.python-version }}
           path: dist/
 
       - name: Install wheel and test dependencies
@@ -372,10 +384,22 @@ jobs:
           name: wheels-linux-cuda
           path: dist/
 
-      - name: Download Windows CUDA wheel
+      - name: Download Windows CUDA wheel (3.10)
+        uses: actions/download-artifact@v4
+        with:
+          name: wheels-windows-cuda-3.10
+          path: dist/
+
+      - name: Download Windows CUDA wheel (3.11)
+        uses: actions/download-artifact@v4
+        with:
+          name: wheels-windows-cuda-3.11
+          path: dist/
+
+      - name: Download Windows CUDA wheel (3.12)
         uses: actions/download-artifact@v4
         with:
-          name: wheels-windows-cuda
+          name: wheels-windows-cuda-3.12
           path: dist/
 
       - name: Download CPU-only wheel

comfy_kitchen/backends/cuda/CMakeLists.txt

@@ -170,11 +170,20 @@ if(CMAKE_CONFIGURATION_TYPES)
     )
 endif()
 
-# Set properties for ABI3 compatibility (Stable ABI)
-target_compile_definitions(_C PRIVATE
-    Py_LIMITED_API=0x030C0000
-    NB_STATIC
-)
+# Set properties for ABI3 compatibility (Stable ABI) - Python 3.12+ only
+# Nanobind's stable ABI support requires Python 3.12+
+if(Python_VERSION VERSION_GREATER_EQUAL "3.12")
+    message(STATUS "Python ${Python_VERSION} detected - enabling stable ABI (abi3)")
+    target_compile_definitions(_C PRIVATE
+        Py_LIMITED_API=0x030C0000
+        NB_STATIC
+    )
+else()
+    message(STATUS "Python ${Python_VERSION} detected - building version-specific wheel")
+    target_compile_definitions(_C PRIVATE
+        NB_STATIC
+    )
+endif()
 
 
 # Install target

comfy_kitchen/tensor/base.py

@@ -7,7 +7,7 @@
 from abc import ABC, abstractmethod
 from dataclasses import dataclass
 from functools import lru_cache
-from typing import Any, Self
+from typing import Any
 
 import torch
 import torch._dynamo
@@ -39,21 +39,21 @@ def _tensor_fields(self) -> list[str]:
         """Return list of field names that are tensors. Override in subclass."""
         return ["scale"]
 
-    def to_device(self, device: torch.device) -> Self:
+    def to_device(self, device: torch.device) -> BaseLayoutParams:
         """Move all tensor fields to the specified device."""
         kwargs = {f.name: getattr(self, f.name) for f in dataclasses.fields(self)}
         for field in self._tensor_fields():
             kwargs[field] = kwargs[field].to(device=device)
         return type(self)(**kwargs)
 
-    def clone(self) -> Self:
+    def clone(self) -> BaseLayoutParams:
         """Clone all tensor fields."""
         kwargs = {f.name: getattr(self, f.name) for f in dataclasses.fields(self)}
         for field in self._tensor_fields():
             kwargs[field] = kwargs[field].clone()
         return type(self)(**kwargs)
 
-    def copy_from(self, src: Self, non_blocking: bool = False) -> None:
+    def copy_from(self, src: BaseLayoutParams, non_blocking: bool = False) -> None:
         """Copy tensor fields in-place from src, reusing existing memory.
 
         Args:

pyproject.toml

@@ -3,15 +3,16 @@ requires = [
     "setuptools>=61.0",
     "wheel",
     "nanobind>=2.0.0",
+    "tomli; python_version < '3.11'",
 ]
 build-backend = "setuptools.build_meta"
 
 [project]
 name = "comfy-kitchen"
-version = "0.1.1"
+version = "0.1.2"
 description = "Fast Kernel Library for ComfyUI with multiple compute backends"
 readme = "README.md"
-requires-python = ">=3.12"
+requires-python = ">=3.10"
 license = {text = "Apache-2.0"}
 keywords = ["pytorch", "cuda", "quantization", "kernels", "diffusion"]
 classifiers = [
@@ -22,8 +23,8 @@ classifiers = [
 ]
 dependencies = [
     "torch>=2.5.0",
+    "nvidia-cublas>=13.0.0",
 ]
-# "nvidia-cublas>=13.0.0", <= we are purposefully leaving this out for now, should be installed when NVIDIA pytorch is
 
 [project.urls]
 Homepage = "https://github.com/comfyanonymous/ComfyKitchen"
@@ -50,8 +51,8 @@ exclude = ["tests*"]
 # Set the maximum line length to 100.
 line-length = 100
 
-# Assume Python 3.12
-target-version = "py312"
+# Assume Python 3.10 (lowest supported)
+target-version = "py310"
 
 # Exclude commonly ignored directories
 exclude = [

setup.py

@@ -4,9 +4,13 @@
 import shutil
 import subprocess
 import sys
-import tomllib
 from typing import ClassVar
 
+try:
+    import tomllib
+except ModuleNotFoundError:
+    import tomli as tomllib
+
 import setuptools
 from setuptools import Extension
 from setuptools.command.build_ext import build_ext
@@ -289,9 +293,9 @@ def get_cmdclass(has_extensions):
         class CUDABdistWheel(bdist_wheel):
             def finalize_options(self):
                 super().finalize_options()
-                if not BUILD_NO_CUDA:
-                    # Set stable ABI tag: cp312-abi3 instead of cp312-cp312
-                    # This indicates the extension uses Python's Limited API
+                # Set stable ABI tag only for Python 3.12+ (nanobind requirement)
+                # For 3.10/3.11, leave as version-specific (cpXXX-cpXXX)
+                if not BUILD_NO_CUDA and sys.version_info >= (3, 12):
                     self.py_limited_api = "cp312"
 
         cmdclass["bdist_wheel"] = CUDABdistWheel