Fix hash collision for multi modal models (#2256)
This commit is contained in:
@@ -216,6 +216,7 @@ class TokenizerManager:
|
||||
input_ids = obj.input_ids
|
||||
|
||||
if self.is_generation:
|
||||
# TODO: also support getting embeddings for multimodal models
|
||||
image_inputs: Dict = await self.image_processor.process_images_async(
|
||||
obj.image_data, input_text or input_ids, obj
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user