Update modeling_minicpmv.py

modeling_minicpmv.py

@@ -186,7 +186,7 @@ class MiniCPMV(MiniCPMVPreTrainedModel):
         terminators = None
         if tokenizer is not None:
             terminators = [tokenizer.convert_tokens_to_ids(i) for i in self.terminators]
-        kwargs.pop("image_sizes")
+        kwargs.pop("image_sizes", None)
         output = self.llm.generate(
             inputs_embeds=inputs_embeds,
             # pad_token_id=0,
@@ -392,7 +392,7 @@ class MiniCPMV(MiniCPMVPreTrainedModel):
 
         generation_config.update((k, kwargs[k]) for k in generation_config.keys() & kwargs.keys())
 
-        inputs.pop("image_sizes")
+        inputs.pop("image_sizes", None)
         with torch.inference_mode():
             res = self.generate(
                 **inputs,