[Bugfix] Update InternVL input mapper to support image embeds (#9351)

vllm-project · Oct 15, 2024 · 55e081f · 55e081f
1 parent 8e836d9
commit 55e081f
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/vllm/model_executor/models/internvl.py b/vllm/model_executor/models/internvl.py
@@ -342,6 +342,8 @@ def input_mapper(
         elif is_list_of(data, Image.Image):
             # we can't stack here because images may have different num_patches
             data = [image_pixel_values_mapper(img) for img in data]
+        else:
+            return MultiModalInputs({"image_embeds": data})
         model_config = ctx.model_config
         tokenizer = cached_get_tokenizer(
             model_config.tokenizer,