EngineX-Iluvatar/enginex-bi_series-vllm
16
1
Fork 0
Code Issues 1 Pull Requests Actions Projects Releases Wiki Activity

Update to new version of base image

Browse Source
This commit is contained in:
Li Jiashu
2025-10-24 15:45:06 +08:00
parent ee04aead1e
commit fad74b701b
476 changed files with 1270 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/qwen3.py
View File

@@ -136,11 +136,11 @@ class Qwen3Attention(nn.Module):
# Add qk-norm
q_by_head = q.view(*q.shape[:-1], q.shape[-1] // self.head_dim,
self.head_dim)
q_by_head = self.q_norm.forward_native(q_by_head)
q_by_head = self.q_norm.forward_cuda(q_by_head.contiguous())
q = q_by_head.view(q.shape)
k_by_head = k.view(*k.shape[:-1], k.shape[-1] // self.head_dim,
self.head_dim)
k_by_head = self.k_norm.forward_native(k_by_head)
k_by_head = self.k_norm.forward_cuda(k_by_head.contiguous())
k = k_by_head.view(k.shape)
q, k = self.rotary_emb(positions, q, k)
attn_output = self.attn(q, k, v, kv_cache, attn_metadata)