[BugFix] Fix vllm_flash_attn install issues (#17267)

Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
Co-authored-by: Aaron Pham <contact@aarnphm.xyz>
This commit is contained in:
Lucas Wilkinson
2025-04-27 17:27:56 -07:00
committed by GitHub
co-authored by Jee Jee Li Aaron Pham
parent 20e489eaa1
commit d8bccde686
11 changed files with 28 additions and 284 deletions
+1 -1
View File
@@ -205,6 +205,7 @@ from vllm.attention.backends.utils import (PAD_SLOT_ID, compute_slot_mapping,
compute_slot_mapping_start_idx,
is_block_tables_empty)
from vllm.attention.ops.merge_attn_states import merge_attn_states
from vllm.attention.utils.fa_utils import get_flash_attn_version
from vllm.model_executor.layers.linear import (ColumnParallelLinear,
LinearBase, RowParallelLinear,
UnquantizedLinearMethod)
@@ -214,7 +215,6 @@ from vllm.multimodal import MultiModalPlaceholderMap
from vllm.platforms import current_platform
from vllm.triton_utils import HAS_TRITON
from vllm.utils import async_tensor_h2d, cdiv, make_tensor_with_pad, round_down
from vllm.vllm_flash_attn.fa_utils import get_flash_attn_version
if HAS_TRITON:
from vllm.attention.ops.triton_flash_attention import triton_attention