biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Frontend] Support multimodal inputs for late-interaction scoring (ColQwen3) + NewModel: nvidia/nemotron-colembed (#34574)

Browse Source 
Signed-off-by: craftsangjae <craftsangjae@gmail.com>
This commit is contained in:
Kata Coder
2026-02-21 13:01:40 +09:00
committed by GitHub
parent 11be2c74dc
commit 5719a4e4e6
10 changed files with 532 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
vllm/model_executor/models/registry.py
View File

@@ -256,6 +256,7 @@ _EMBEDDING_MODELS = {
"Qwen2VLForConditionalGeneration": ("qwen2_vl", "Qwen2VLForConditionalGeneration"), # noqa: E501
"ColQwen3": ("colqwen3", "ColQwen3Model"),
"OpsColQwen3Model": ("colqwen3", "ColQwen3Model"),
"Qwen3VLNemotronEmbedModel": ("colqwen3", "ColQwen3Model"),
"SiglipModel": ("siglip", "SiglipEmbeddingModel"),
# Technically Terratorch models work on images, both in
# input and output. I am adding it here because it piggy-backs on embedding