xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-21 02:15:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Model][Qwen3VL] Slighly speedup fast_pos_embed_interpolate (#28434)

Browse Source 
Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com>
This commit is contained in:
Lukas Geiger 2025-11-11 10:35:10 +00:00 committed by GitHub
parent c7991269dd
commit 9973e6e04a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/qwen3_vl.py
View File

@ -491,8 +491,8 @@ class Qwen3_VisionTransformer(nn.Module):
weights = weights.to(dtype=self.dtype) weights = weights.to(dtype=self.dtype)
embeds = self.pos_embed(indices) embeds = self.pos_embed(indices)
weighted_embeds = embeds * weights embeds *= weights
combined = weighted_embeds.sum(dim=0) combined = embeds.sum(dim=0)
combined = combined.reshape( combined = combined.reshape(
h // m_size, m_size, w // m_size, m_size, hidden_dim h // m_size, m_size, w // m_size, m_size, hidden_dim