Update launch.py

launch.py

@@ -33,7 +33,7 @@ lm_model = LlamaForCausalLM.from_pretrained(
     device_map="auto",
     torch_dtype=torch.float16,
     use_auth_token=AUTH_TOKEN,
-    quantization_config=BitsAndBytesConfig(load_in_8bit=True))
+    quantization_config=BitsAndBytesConfig(load_in_4bit=True))
 tokenizer = LlamaTokenizer.from_pretrained(LM_MODEL_PATH, use_auth_token=AUTH_TOKEN)
 
 # visual model