Skip to content

GitLab

Unverified Commit 9973e6e0 authored by Lukas Geiger's avatar Lukas Geiger Committed by GitHub
Browse files

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


Signed-off-by: default avatarLukas Geiger <lukas.geiger94@gmail.com>
parent c7991269
Hide whitespace changes
Inline Side-by-side
Showing with 2 additions and 2 deletions
vllm/model_executor/models/qwen3_vl.py
View file @ 9973e6e0
...@@ -491,8 +491,8 @@ class Qwen3_VisionTransformer(nn.Module): ...@@ -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
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment