Fix

app.py

@@ -58,6 +58,19 @@ if __name__ == "__main__":
     gws.args = parser.parse_args()
     gws.models = []
 
+    gws.title_markdown += """
+
+ONLY WORKS WITH GPU! By default, we load the model with 4-bit quantization to make it fit in smaller hardwares. Set the environment variable `bits` to control the quantization.
+
+Set the environment variable `model` to change the model, and switch hardware accordingly:
+| Model | Hardware          |
+|-------|-------------------|
+| liuhaotian/llava-v1.6-mistral-7b  | T4-medium |
+| liuhaotian/llava-v1.6-vicuna-7b   | T4-medium |
+| liuhaotian/llava-v1.6-vicuna-13b  | T4-medium |
+| liuhaotian/llava-v1.6-34b      | 2xA10G large |
+"""
+
     print(f"args: {gws.args}")
 
     model_path = os.getenv("model", "liuhaotian/llava-v1.6-mistral-7b")