Merge branch 'main' of https://huggingface.co/spaces/OpenGVLab/VideoChatGPT

README.md

@@ -1,5 +1,5 @@
 ---
-title: Ask-Anything:ChatGPT with Video Understanding
+title: 'VideoChat: Chat-Centric Video Understanding'
 emoji: 👀
 colorFrom: green
 colorTo: blue

configs/config.json

@@ -3,7 +3,7 @@
     "vit_model": "eva_clip_g",
     "vit_model_path": "model/eva_vit_g.pth",
     "q_former_model_path": "model/blip2_pretrained_flant5xxl.pth",
-    "llama_model_path": "model/stable-vicuna-13b",
+    "llama_model_path": "ynhe/stable-vicuna-13b",
     "videochat_model_path": "model/videochat.pth",
     "img_size": 224,
     "num_query_token": 32,

models/videochat.py

@@ -1,3 +1,4 @@
+import os
 import random
 import logging
 
@@ -109,7 +110,7 @@ class VideoChat(Blip2Base):
         print('Loading Q-Former Done')
 
         print('Loading LLAMA')
-        self.llama_tokenizer = LlamaTokenizer.from_pretrained(llama_model_path, use_fast=False)
+        self.llama_tokenizer = LlamaTokenizer.from_pretrained(llama_model_path, use_fast=False, use_auth_token=os.environ["HF_TOKEN"])
         self.llama_tokenizer.pad_token = self.llama_tokenizer.eos_token
 
         if self.low_resource:
@@ -117,12 +118,14 @@ class VideoChat(Blip2Base):
                     llama_model_path,
                     torch_dtype=torch.float16,
                     load_in_8bit=True,
-                    device_map="auto"
+                    device_map="auto",
+                    use_auth_token=os.environ["HF_TOKEN"],
                 )
         else:
                 self.llama_model = LlamaForCausalLM.from_pretrained(
                     llama_model_path,
                     torch_dtype=torch.float16,
+                    use_auth_token=os.environ["HF_TOKEN"],
                 )
 
         print("freeze LLAMA")