Create app.py

app.py

@@ -0,0 +1,48 @@
+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)
+
+# Load the processor
+processor = AutoProcessor.from_pretrained("Salesforce/blip2-opt-2.7b")
+
+# Define the prediction function
+def predict(image):
+    # Preprocess the image
+    inputs = processor(images=image, return_tensors="pt").to(device, torch.float16)
+    new_eos_token_id = 13
+    with torch.no_grad():
+        generated_ids = modelA.generate(**inputs, max_length=100,
+                                      eos_token_id=new_eos_token_id)
+    generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)
+    return f"{generated_caption[0]}"
+
+# Set up the Gradio interface
+demo = gr.Interface(
+    fn=predict,
+    inputs=gr.inputs.Image(type="pil"),  # Upload an image in PIL format
+    outputs="text",  # The output will be the generated caption
+    title="Strep Throat Image Assessment",
+    description="Upload an image of a throat and receive a medical assessment caption based on the model's output."
+)
+
+# Launch the Gradio app
+demo.launch()