Update app.py

app.py

@@ -3,16 +3,20 @@ from transformers import AutoProcessor, PaliGemmaForConditionalGeneration, BitsA
 from peft import PeftModel
 import spaces
 import torch
+from huggingface_hub.hf_api import HfFolder
+import os
+token = os.getenv('token')
+HfFolder.save(token)
 device = "cuda"
-base_model = PaliGemmaForConditionalGeneration.from_pretrained("google/paligemma-3b-mix-224").to(device)
+base_model = PaliGemmaForConditionalGeneration.from_pretrained("google/paligemma-3b-mix-224")
 processor = AutoProcessor.from_pretrained("google/paligemma-3b-mix-224")
 
 @spaces.GPU(duration=120)
 def greet(image, prompt):
     
-    quantization_config = BitsAndBytesConfig(
-        load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.float16
-    )
+    # quantization_config = BitsAndBytesConfig(
+    #     load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.float16
+    # )
     
     
     # model = PaliGemmaForConditionalGeneration.from_pretrained("/folders", torch_dtype=torch.float16, quantization_config=quantization_config).to(device)