Update app.py

app.py

@@ -1,29 +1,24 @@
 import torch
 import gradio as gr
-from transformers import Blip2ForConditionalGeneration, AutoProcessor, BitsAndBytesConfig
-from peft import PeftModel, PeftConfig
-
-# Load the PEFT model configuration and quantization settings
-peft_model_id = "Prasi21/blip2-opt-2.7b-strep-throat-caption-adapters3"
-config = PeftConfig.from_pretrained(peft_model_id)
-config.base_model_name_or_path = "Prasi21/blip2-opt-2.7b-strep-throat-caption-adapters3"
-
-# Enable 8-bit quantization for more efficient loading
-quantization_config = BitsAndBytesConfig(load_in_8bit=True)
-
-# Load the base model with quantization
-model = Blip2ForConditionalGeneration.from_pretrained(
-    config.base_model_name_or_path,
-    quantization_config=quantization_config,
-    device_map="auto"
-)
-
-# Load the fine-tuned PEFT model
-model = PeftModel.from_pretrained(model, peft_model_id)
+from transformers import AutoProcessor, Blip2ForConditionalGeneration
+from peft import LoraConfig, get_peft_model, PeftModel
 
 # Load the processor
 processor = AutoProcessor.from_pretrained("Salesforce/blip2-opt-2.7b")
 
+# Load the base model from the original repository
+base_model = Blip2ForConditionalGeneration.from_pretrained(
+    "ybelkada/blip2-opt-2.7b-fp16-sharded",
+    device_map="auto",
+    quantization_config=quantization_config
+)
+
+repo_id = "Prasi21/blip2-opt-2.7b-strep-throat-caption-adapters"
+
+# Load the fine-tuned LoRA adapters from the Hugging Face Hub
+model = PeftModel.from_pretrained(base_model, repo_id)
+
+
 # Define the prediction function
 def predict(image):
     # Preprocess the image