Update llava/model/llava_arch.py

llava/model/llava_arch.py

@@ -773,7 +773,7 @@ class LlavaMetaForCausalLM(ABC):
         **generation_kwargs,
     ):
         inputs_embeds, _, attention_mask = self._embed(input_ids, media, media_config, None, attention_mask, media_meta)
-        return self.llm.generate(inputs_embeds=inputs_embeds, attention_mask=attention_mask, **generation_kwargs)
+        return self.llm.generate(inputs_embeds=inputs_embeds, attention_mask=attention_mask, max_new_tokens=1024, **generation_kwargs)
 
     @torch.inference_mode()
     def generate_content(