Upload moondream.py

moondream.py

@@ -321,13 +321,15 @@ class Moondream(PreTrainedModel):
             **kwargs,
     ):
         prompt = f"<image>\n\n{chat_history}Question: {question}\n\nAnswer:"
-        return self.generate(
+        generator = self.generate(
             image_embeds,
             prompt,
             tokenizer=tokenizer,
             max_new_tokens=512,
             **kwargs,
         )
+        async for token in generator:
+            yield token
 
     def batch_answer(
         self,