xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 23:34:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BUGFIX] deepseek-v2-lite failed due to fused_qkv_a_proj name update (#21414)

Browse Source 
Signed-off-by: Chendi.Xue <chendi.xue@intel.com>
This commit is contained in:
Chendi.Xue 2025-07-22 22:33:57 -05:00 committed by GitHub
parent 4f76a05f4f
commit 08d2bd78da
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/model_executor/models/deepseek_v2.py
View File

@ -885,13 +885,16 @@ class DeepseekV2ForCausalLM(nn.Module, SupportsPP, MixtureOfExperts):
# for mlp.experts[0].gate_gate_up_proj, which breaks load.
if (("mlp.experts." in name) and name not in params_dict):
continue
name = name.replace(weight_name, param_name)
name_mapped = name.replace(weight_name, param_name)
# QKV fusion is optional, fall back to normal
# weight loading if it's not enabled
# if go with fusion option, then update name
if ((param_name == "fused_qkv_a_proj")
and name not in params_dict):
and name_mapped not in params_dict):
continue
else:
name = name_mapped
# Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict:
continue