Update minigpt4/models/mini_gpt4.py

minigpt4/models/mini_gpt4.py

@@ -38,7 +38,7 @@ class MiniGPT4(Blip2Base):
         freeze_vit=True,
         freeze_qformer=True,
         num_query_token=32,
-        llama_model="/data/vicuna-7b",
+        llama_model="",
         llama_cache_dir='',
         prompt_path="",
         prompt_template="",
@@ -87,16 +87,16 @@ class MiniGPT4(Blip2Base):
         print('Loading Q-Former Done')
 
         print('Loading LLAMA')
-        self.llama_tokenizer = LlamaTokenizer.from_pretrained('/data/vicuna-7b', use_fast=False, use_auth_token=True)
+        self.llama_tokenizer = LlamaTokenizer.from_pretrained('Vision-CAIR/vicuna-7b', use_fast=False, use_auth_token=True)
         self.llama_tokenizer.pad_token = self.llama_tokenizer.eos_token
 
         if llama_cache_dir:
             self.llama_model = LlamaForCausalLM.from_pretrained(
-                '/data/vicuna-7b', load_in_4bit=True, torch_dtype=torch.float16, device_map="auto", use_auth_token=True
+                'Vision-CAIR/vicuna-7b', load_in_4bit=True, torch_dtype=torch.float16, device_map="auto", use_auth_token=True
             )
         else:
             self.llama_model = LlamaForCausalLM.from_pretrained(
-                '/data/vicuna-7b', load_in_4bit=True, torch_dtype=torch.float16, device_map="auto", use_auth_token=True
+                'Vision-CAIR/vicuna-7b', load_in_4bit=True, torch_dtype=torch.float16, device_map="auto", use_auth_token=True
             )
         for name, param in self.llama_model.named_parameters():
             param.requires_grad = False