xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-25 09:16:32 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bug] Fix Number of dimensions of tensors must match. for Deepseek V3.2 (#31160)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye 2025-12-23 21:41:09 -05:00 committed by GitHub
parent 8b59753cdb
commit 76e6a95192
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/models/deepseek_v2.py
View File

@ -878,11 +878,14 @@ class Indexer(nn.Module):
)
q_pe, k_pe = rotary_emb(positions, q_pe, k_pe.unsqueeze(1))
# `rotary_emb` is shape-preserving; `q_pe` is already
# [num_tokens, n_head, rope_dim].
# Note: RoPE (NeoX) can introduce extra leading dimensions during compilation
# so we need to reshape back to token-flattened shapes
q_pe = q_pe.reshape(-1, self.n_head, self.rope_dim)
k_pe = k_pe.reshape(-1, 1, self.rope_dim)
q = torch.cat([q_pe, q_nope], dim=-1)
# `k_pe` is [num_tokens, 1, rope_dim] (MQA).
k = torch.cat([k_pe.squeeze(1), k_nope], dim=-1)
k = torch.cat([k_pe.squeeze(-2), k_nope], dim=-1)
# we only quant q here since k quant is fused with cache insertion
q = q.view(-1, self.head_dim)