obscura/vllm
2
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2026-06-06 00:16:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

[FlashAttention] Don't overwrite flash_attn_interface.py when installing precompiled (#39932)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni
2026-04-15 16:43:15 -04:00
committed by GitHub
parent ac3dac545b
commit c77e596e2e
1 changed files with 11 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
setup.py
+11 -1
View File
@@ -693,6 +693,12 @@ class precompiled_wheel_utils:
flash_attn_regex = re.compile(
r"vllm/vllm_flash_attn/(?:[^/.][^/]*/)*(?!\.)[^/]*\.py"
)
# __init__.py and flash_attn_interface.py are source-controlled
# in vllm and should not be overwritten (matches cmake exclusions)
flash_attn_files_to_skip = {
"vllm/vllm_flash_attn/__init__.py",
"vllm/vllm_flash_attn/flash_attn_interface.py",
}
triton_kernels_regex = re.compile(
r"vllm/third_party/triton_kernels/(?:[^/.][^/]*/)*(?!\.)[^/]*\.py"
)
@@ -705,7 +711,11 @@ class precompiled_wheel_utils:
filter(lambda x: x.filename in files_to_copy, wheel.filelist)
)
file_members += list(
filter(lambda x: flash_attn_regex.match(x.filename), wheel.filelist)
filter(
lambda x: flash_attn_regex.match(x.filename)
and x.filename not in flash_attn_files_to_skip,
wheel.filelist,
)
)
file_members += list(
filter(