set model=Tarsier2-7b

app.py

@@ -23,6 +23,8 @@ from dataset.processor import Processor
 import os
 import torch
 
+# huggingface-cli login
+
 device = 'cuda'
 model_path = os.getenv("MODEL_PATH", "omni-research/Tarsier2-7b")
 max_n_frames = int(os.getenv("MAX_N_FRAMES", 8))

dataset/processor.py

@@ -86,7 +86,7 @@ class Processor(object):
             model_name_or_path,
             padding_side='left',
             trust_remote_code=True,
-            token=HF_TOKEN,
+            # token=HF_TOKEN,
         )
         self.processor = CustomImageProcessor(sub_processor)
         self.tokenizer = sub_processor.tokenizer

tools/utils.py

@@ -55,7 +55,7 @@ def load_model_and_processor(model_name_or_path, max_n_frames=8):
     model_config = LlavaConfig.from_pretrained(
         model_name_or_path,
         trust_remote_code=True,
-        token=HF_TOKEN,
+        # token=HF_TOKEN,
     )
     model = TarsierForConditionalGeneration.from_pretrained(
         model_name_or_path,
@@ -63,7 +63,7 @@ def load_model_and_processor(model_name_or_path, max_n_frames=8):
         device_map='auto',
         torch_dtype=torch.float16,
         trust_remote_code=True,
-        token=HF_TOKEN,
+        # token=HF_TOKEN,
     )
     model.eval()
     return model, processor