update

bubogpt/models/mm_gpt4.py

@@ -84,12 +84,11 @@ class MMGPT4(BaseModel):
         print('Loading ImageBind Done')
 
         print(f'Loading LLAMA from {llama_model}')
-        self.llama_tokenizer = LlamaTokenizer.from_pretrained(llama_model, use_fast=False)
+        self.llama_tokenizer = LlamaTokenizer.from_pretrained('magicr/vicuna-7b', use_fast=False, use_auth_token=True)
         self.llama_tokenizer.pad_token = self.llama_tokenizer.eos_token
 
         self.llama_model = LlamaForCausalLM.from_pretrained(
-            llama_model,
-            torch_dtype=torch.float16,
+            'magicr/vicuna-7b', load_in_8bit=True, torch_dtype=torch.float16, device_map="auto", use_auth_token=True
         )
 
         if freeze_llm: