Feature/vit attention unification# 23880 (#23978)

Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
2025-09-10 09:10:14 -04:00
parent 72d30108a0
commit 6cbd41909e
9 changed files with 68 additions and 56 deletions
--- a/vllm/model_executor/models/vision.py
+++ b/vllm/model_executor/models/vision.py
@@ -122,4 +122,4 @@ def resolve_visual_encoder_outputs(
    uses_last_layer = feature_sample_layers[-1] in (len(hs_pool) - 1, -1)
    if post_layer_norm is not None and uses_last_layer:
        hs_pool[-1] = post_layer_norm(encoder_outputs)
-    return torch.cat(hs_pool, dim=-1)
+    return torch.cat(hs_pool, dim=-1)