Build older cuda wheels (#158)

.github/workflows/build.yaml

@@ -41,11 +41,12 @@ jobs:
       matrix:
         os: [ubuntu-20.04, windows-latest]
         pyver: ["3.8", "3.9", "3.10", "3.11"]
-        cuda: ["12.1.1"]
+        cuda: ["11.8.0", "12.1.1"]
     defaults:
       run:
         shell: pwsh
     env:
+      PYPI_CUDA_VERSION: "12.1.1"  
       CUDA_VERSION: ${{ matrix.cuda }}
     
     steps:
@@ -105,7 +106,11 @@ jobs:
         run: |
           $env:CUDA_PATH = $env:CONDA_PREFIX
           $env:CUDA_HOME = $env:CONDA_PREFIX
+
+          # Only add +cu118 to wheel if not releasing on PyPi
+          if ( $env:CUDA_VERSION -eq $env:PYPI_CUDA_VERSION ){
             $env:PYPI_BUILD = 1
+          }
 
           python setup.py sdist bdist_wheel
       

README.md

@@ -34,12 +34,23 @@ Requirements:
 ---
 
 Install:
-- Use pip to install awq
+
+- Install from PyPi distributed wheels (torch 2.1.0 + CUDA 12.1.1)
 
 ```
 pip install autoawq
 ```
 
+- Install from GitHub a release (torch 2.0.1 + CUDA 11.8.0)
+
+Remember to grab the right link for the [latest release](https://github.com/casper-hansen/AutoAWQ/releases) that matches your environment.
+
+For example, this wheel is torch 2.0.1 with CUDA 11.8.0 and Python 3.10 for Linux:
+
+```
+pip install https://github.com/casper-hansen/AutoAWQ/releases/download/v0.1.6/autoawq-0.1.6+cu118-cp310-cp310-linux_x86_64.whl
+```
+
 ### Using conda
 
 CUDA dependencies can be hard to manage sometimes. It is recommended to use conda with AutoAWQ:

setup.py

@@ -43,7 +43,7 @@ common_setup_kwargs = {
 }
 
 requirements = [
-    "torch>=2.1.0",
+    "torch>=2.0.1",
     "transformers>=4.35.0",
     "tokenizers>=0.12.1",
     "accelerate",