frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use hidden_size_per_head as head_size fallback (#24221)

Browse Source 
Signed-off-by: nopperl <54780682+nopperl@users.noreply.github.com>
This commit is contained in:
nopperl
2025-09-04 20:59:16 +09:00
committed by GitHub
parent eafa8dcde6
commit 2b30afa442
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/config/__init__.py
View File

@ -1426,6 +1426,11 @@ class ModelConfig:
if getattr(self.hf_text_config, "head_dim", None) is not None:
return self.hf_text_config.head_dim
# NOTE: Some models (such as PLaMo2.1) use `hidden_size_per_head`
if getattr(self.hf_text_config, "hidden_size_per_head",
None) is not None:
return self.hf_text_config.hidden_size_per_head
# FIXME(woosuk): This may not be true for all models.
return (self.hf_text_config.hidden_size //
self.hf_text_config.num_attention_heads)