Re-disable dropout and alibi (not currently implemented) to save binary size.

Author: nshepperd

CMakeLists.txt

@@ -34,11 +34,6 @@ if(CCACHE_PROGRAM)
     message(STATUS "Using ccache: ${CCACHE_PROGRAM}")
 endif()
 
-# Options from environment variables
-option(FLASH_ATTENTION_FORCE_BUILD "Force building from source" OFF)
-option(FLASH_ATTENTION_SKIP_CUDA_BUILD "Skip CUDA build" OFF)
-option(FLASH_ATTENTION_FORCE_CXX11_ABI "Force using C++11 ABI" OFF)
-
 # CUDA handling
 # Get CUDA architectures from environment or use default
 if(DEFINED ENV{FLASH_ATTN_CUDA_ARCHS})
@@ -62,6 +57,8 @@ set(CUDA_FLAGS
     -U__CUDA_NO_HALF_CONVERSIONS__
     -U__CUDA_NO_HALF2_OPERATORS__
     -U__CUDA_NO_BFLOAT16_CONVERSIONS__
+    -DFLASHATTENTION_DISABLE_DROPOUT=1
+    -DFLASHATTENTION_DISABLE_ALIBI=1
 )
 
 # Collect source files

pyproject.toml

@@ -60,7 +60,7 @@ provider = "scikit_build_core.metadata.regex"
 input = "src/flash_attn_jax/__init__.py"
 
 [tool.cibuildwheel]
-manylinux-x86_64-image = "quay.io/pypa/manylinux_2_28_x86_64:latest"
+# manylinux-x86_64-image = "quay.io/pypa/manylinux_2_28_x86_64:latest"
 before-all = "bash scripts/install-cuda-linux.sh"
 build = "cp312-manylinux_x86_64"
 repair-wheel-command = "auditwheel repair --exclude=libcudart.so* -w {dest_dir} {wheel}"