Update for peft_model

app.py

@@ -35,7 +35,9 @@ def load_model_and_adapters():
     )
     print("Loading your fine-tuned fridge adapters...")
     # PEFT will look at base_model's device map to assign tensor locations safely
-    peft_model = PeftModel.from_pretrained(base, "uttarasawant/qwen2.5-vl-fridge-adapters")
+    #peft_model = PeftModel.from_pretrained(base, "uttarasawant/qwen2.5-vl-fridge-adapters")
+    # Force safetensors to read the file into CPU memory first, bypassing the boot-time CUDA check
+    peft_model = PeftModel.from_pretrained(base, "uttarasawant/qwen2.5-vl-fridge-adapters", low_cpu_mem_usage=True)
     
     return peft_model