update

bubogpt/configs/models/mmgpt4.yaml

@@ -10,7 +10,7 @@ model:
   num_query_token: 32
 
   # Vicuna
-  llama_model: "vicuna"
+  llama_model: "magicr/vicuna-7b"
 
   # generation configs
   prompt: ""

bubogpt/models/mm_gpt4.py

@@ -86,10 +86,10 @@ class MMGPT4(BaseModel):
         gc.collect()
 
         print(f'Loading LLAMA from {llama_model}')
-        self.llama_tokenizer = LlamaTokenizer.from_pretrained('magicr/vicuna-7b', use_fast=False, use_auth_token=True)
+        self.llama_tokenizer = LlamaTokenizer.from_pretrained(llama_model, use_fast=False, use_auth_token=True)
         self.llama_tokenizer.pad_token = self.llama_tokenizer.eos_token
 
-        self.llama_model = LlamaForCausalLM.from_pretrained('magicr/vicuna-7b', load_in_8bit=True,
+        self.llama_model = LlamaForCausalLM.from_pretrained(llama_model, load_in_8bit=True,
                                                             torch_dtype=torch.float16, device_map="auto", use_auth_token=True)
 
         if freeze_llm: