biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Misc] Use apply_rotary_emb from vllm_flash_attn for Qwen2-VL vision RoPE (#17726)

Browse Source 
Signed-off-by: Isotr0py <2037008807@qq.com>
This commit is contained in:
Isotr0py
2025-05-07 12:51:33 +08:00
committed by GitHub
parent 822de7fb94
commit c3e9d5060e
2 changed files with 6 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/models/qwen2_5_vl.py
View File

@@ -297,13 +297,8 @@ class Qwen2_5_VisionAttention(nn.Module):
q, k, v = (rearrange(x, "s b ... -> b s ...").contiguous()
for x in (q, k, v))
if rotary_pos_emb is not None:
use_flash_attn = self.attn_backend == _Backend.FLASH_ATTN
q = apply_rotary_pos_emb_vision(q,
rotary_pos_emb,
use_flash_attn=use_flash_attn)
k = apply_rotary_pos_emb_vision(k,
rotary_pos_emb,
use_flash_attn=use_flash_attn)
q = apply_rotary_pos_emb_vision(q, rotary_pos_emb)
k = apply_rotary_pos_emb_vision(k, rotary_pos_emb)
if self.attn_backend == _Backend.FLASH_ATTN:
# from vllm_flash_attn.flash_attn_interface import (