Create app.py

app.py

@@ -0,0 +1,78 @@
+import gradio as gr
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+
+# 1. Define your model ID
+# REPLACE THIS with your actual username/repo name
+ADAPTER_ID = "shri171981/medical_chat_generative" 
+
+def load_model():
+    # Load Base Model (Llama-3-8B)
+    # We use "cpu" and float32 if you are on the Free Tier (Slow but works)
+    # If you have a GPU in your Space, change device_map to "auto"
+    base_model_name = "unsloth/llama-3-8b-instruct-bnb-4bit"
+    
+    print("Loading base model...")
+    base_model = AutoModelForCausalLM.from_pretrained(
+        base_model_name,
+        device_map="cpu", # Change to "auto" if you have a GPU Space
+        torch_dtype=torch.float32,
+        low_cpu_mem_usage=True
+    )
+    
+    print("Loading adapter...")
+    model = PeftModel.from_pretrained(base_model, ADAPTER_ID)
+    tokenizer = AutoTokenizer.from_pretrained(base_model_name)
+    return model, tokenizer
+
+# Load the model once at startup
+model, tokenizer = load_model()
+
+def ask_doctor(message, history):
+    # 1. Format the input for Llama-3
+    # We strictly enforce the "HACK_DOC" format
+    system_prompt = "You are a helpful and empathetic medical doctor. Answer the patient's question based on the input provided."
+    full_prompt = f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
+
+### Instruction:
+{system_prompt}
+
+### Input:
+{message}
+
+### Response:
+"""
+    
+    # 2. Tokenize and Generate
+    inputs = tokenizer(full_prompt, return_tensors="pt")
+    
+    # Generate response
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs, 
+            max_new_tokens=128,
+            temperature=0.7
+        )
+    
+    # 3. Decode output
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+    # 4. Clean up the text (Remove the prompt part)
+    # We split by "Response:" and take the last part
+    clean_answer = response.split("Response:")[-1].strip()
+    
+    return clean_answer
+
+# 3. Build the UI
+interface = gr.ChatInterface(
+    fn=ask_doctor,
+    title="🚑 HACK_DOC AI",
+    description="I am a specialized medical assistant. Ask me about symptoms!",
+    examples=["I have a sharp pain in my chest.", "What should I take for a fever?", "My skin is itchy and red."],
+    theme="soft"
+)
+
+# 4. Launch
+if __name__ == "__main__":
+    interface.launch()