Update app.py

app.py

@@ -712,11 +712,10 @@ if __name__ == "__main__":
     )
     # argparser.add_argument("--model-path", type=str, default="facebook/opt-350m")
     argparser.add_argument("--model-base", type=str, default=None)
-    argparser.add_argument("--device_map", type=str, default="auto")
     # argparser.add_argument("--num-gpus", type=int, default=1)
     argparser.add_argument("--conv-mode", type=str, default=None)
     argparser.add_argument("--temperature", type=float, default=0.7)
-    argparser.add_argument("--max-new-tokens", type=int, default=4096)
+    argparser.add_argument("--max-new-tokens", type=int, default=250)
     argparser.add_argument("--num_frames", type=int, default=16)
     argparser.add_argument("--load-8bit", action="store_true")
     argparser.add_argument("--load-4bit", action="store_true")
@@ -728,7 +727,8 @@ if __name__ == "__main__":
 
     model_path = args.model_path
     filt_invalid = "cut"
-    lava_kwargs = {"multimodal": True}
+    lava_kwargs = {"multimodal": True, "device_map": "cpu"}
+}
     #model_name = get_model_name_from_path(args.model_path)
     model_name = "pangea_llava_qwen"
     tokenizer, model, image_processor, context_len = load_pretrained_model(model_path, None, model_name, args.load_8bit, args.load_4bit, **lava_kwargs)