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

[CI/Build] Reorganize models tests (#17459)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-05-01 14:03:08 +08:00
committed by GitHub
parent aa4502e7f3
commit afb4429b4f
65 changed files with 316 additions and 323 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/model_executor/models/llama.py
View File

@@ -127,8 +127,10 @@ class LlamaAttention(nn.Module):
assert tp_size % self.total_num_kv_heads == 0
self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)
# MistralConfig has an optional head_dim introduced by Mistral-Nemo
self.head_dim = getattr(config, "head_dim",
self.hidden_size // self.total_num_heads)
head_dim = getattr(config, "head_dim", None)
if head_dim is None:
head_dim = self.hidden_size // self.total_num_heads
self.head_dim = head_dim
# Phi models introduced a partial_rotary_factor parameter in the config
self.partial_rotary_factor = getattr(config, "partial_rotary_factor",
1)