Upload WindowsWhlBuilder_cuda.bat

WindowsWhlBuilder_cuda.bat

@@ -1,9 +1,11 @@
 @echo off
 setlocal enabledelayedexpansion
 
+set MAX_JOBS=1
 
 :parseArgs
-if [%1] == [WORKERS] set MAX_JOBS=%2 & shift & shift & goto :parseargs
+rem Assigning a value to MAX_JOBS via a variable does not work in ninja, I don't know why
+rem if [%1] == [WORKERS] set MAX_JOBS=%2 & shift & shift & goto :parseargs
 if [%1] == [FORCE_CXX11_ABI] set FLASH_ATTENTION_FORCE_CXX11_ABI=%2 & shift & shift & goto :parseargs
 goto :buildContinue
 :end
@@ -27,13 +29,14 @@ rem # We want setuptools >= 49.6.0 otherwise we can't compile the extension if s
 rem # https://github.com/pytorch/pytorch/blob/664058fa83f1d8eede5d66418abff6e20bd76ca8/torch/utils/cpp_extension.py#L810
 rem # However this still fails so I'm using a newer version of setuptools
 rem pip install setuptools==68.0.0
-pip install "setuptools>=49.6.0" ninja packaging wheel psutil
+pip install "setuptools>=49.6.0" packaging wheel psutil
 rem # Limit MAX_JOBS otherwise the github runner goes OOM
 rem # CUDA 11.8 can compile with 2 jobs, but CUDA 12.3 goes OOM
 set FLASH_ATTENTION_FORCE_BUILD=TRUE
 set BUILD_TARGET=cuda
 set DISTUTILS_USE_SDK=1
 set dist_dir=dist
+
 python setup.py bdist_wheel --dist-dir=%dist_dir%