Upload moondream.py

moondream.py

@@ -281,6 +281,7 @@ class Moondream(PreTrainedModel):
 
         #return tokenizer.batch_decode(output_ids, skip_special_tokens=True)
 
+    """
     def answer_question(
         self,
         image_embeds,
@@ -305,6 +306,24 @@ class Moondream(PreTrainedModel):
             result_queue.put(cleaned_answer)
         else:
             return cleaned_answer
+    """
+
+    def answer_question(
+            self,
+            image_embeds,
+            question,
+            tokenizer,
+            chat_history="",
+            **kwargs,
+    ):
+        prompt = f"<image>\n\n{chat_history}Question: {question}\n\nAnswer:"
+        return self.generate(
+            image_embeds,
+            prompt,
+            tokenizer=tokenizer,
+            max_new_tokens=512,
+            **kwargs,
+        )
 
     def batch_answer(
         self,