xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-05-31 19:37:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

Update vllm/attention/backends/mla/utils.py

Browse Source 
Co-authored-by: Michael Goin <mgoin64@gmail.com>
This commit is contained in:
Lucas Wilkinson 2025-01-29 22:51:37 -05:00 committed by GitHub
parent 4880a43d20
commit aa19f297d2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/attention/backends/mla/utils.py
View File

@ -161,7 +161,7 @@ class MLAImplCommon(AttentionImpl):
def _v_up_proj_and_o_proj(self, x): def _v_up_proj_and_o_proj(self, x):
if envs.VLLM_MLA_PERFORM_MATRIX_ABSORPTION: if envs.VLLM_MLA_PERFORM_MATRIX_ABSORPTION:
return self.o_proj_absored( return self.o_proj_absorbed(
x.reshape(-1, self.num_heads * self.kv_lora_rank))[0] x.reshape(-1, self.num_heads * self.kv_lora_rank))[0]
else: else:
x = torch.einsum("bnl,lnv->bnv", x, self.W_UV) x = torch.einsum("bnl,lnv->bnv", x, self.W_UV)