Update app.py

app.py

@@ -1,70 +1,93 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+from peft import PeftModel, PeftConfig # Necessary for loading the adapter weights
 
+# --- Configuration ---
+# 1. Base Llama 2 model used for fine-tuning
+BASE_MODEL = "aboonaji/llama2finetune-v2" 
+# 2. Your newly published adapter model on the Hub
+ADAPTER_MODEL = "dynamodenis254/dynamo-denis-llama2finetune-medical" 
 
-def respond(
-    message,
-    history: list[dict[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-    hf_token: gr.OAuthToken,
-):
-    """
-    For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-    """
-    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
+# --- Model Loading ---
+# This function loads the model and runs only once when the app starts
+def load_model():
+    """Loads the base model and applies the fine-tuned adapter weights."""
+    print(f"Loading base model: {BASE_MODEL}")
+    
+    # Check for GPU availability
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    print(f"Using device: {device}")
 
-    messages = [{"role": "system", "content": system_message}]
+    # Load the base model (ensure trust_remote_code=True for custom Llama models)
+    base_model = AutoModelForCausalLM.from_pretrained(
+        BASE_MODEL,
+        torch_dtype=torch.float16, # Use half precision for faster GPU inference
+        device_map="auto",
+        trust_remote_code=True
+    )
+    
+    # Load the Peft (LoRA) adapter weights on top of the base model
+    model = PeftModel.from_pretrained(base_model, ADAPTER_MODEL)
+    
+    # Get the tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)
+    tokenizer.pad_token = tokenizer.eos_token
+    tokenizer.padding_side = "right"
 
-    messages.extend(history)
+    # Create the Hugging Face Pipeline for easy text generation
+    generator = pipeline(
+        "text-generation",
+        model=model,
+        tokenizer=tokenizer,
+        device=0 if device == "cuda" else -1 # Use GPU 0 if available, otherwise use CPU
+    )
+    
+    print("Model and Tokenizer loaded successfully.")
+    return generator
 
-    messages.append({"role": "user", "content": message})
+# Load the model outside the prediction function so it runs only once
+generator = load_model()
 
-    response = ""
+# --- Prediction Function ---
+def generate_response(prompt, max_new_tokens=256, temperature=0.7):
+    """Generates text using the fine-tuned model."""
+    
+    # Llama models often work best with a system prompt structure
+    system_prompt = "You are a specialized medical assistant. Provide concise and accurate information."
+    formatted_prompt = f"### System:\n{system_prompt}\n\n### User:\n{prompt}\n\n### Assistant:\n"
+    
+    try:
+        # Run the generation pipeline
+        result = generator(
+            formatted_prompt,
+            max_new_tokens=max_new_tokens,
+            temperature=temperature,
+            do_sample=True,
+            return_full_text=False # Only return the generated part of the response
+        )
+        
+        # Extract the text and clean up any potential trailing newlines
+        generated_text = result[0]['generated_text'].strip()
+        return generated_text
+        
+    except Exception as e:
+        return f"An error occurred during generation: {e}"
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        choices = message.choices
-        token = ""
-        if len(choices) and choices[0].delta.content:
-            token = choices[0].delta.content
-
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-chatbot = gr.ChatInterface(
-    respond,
-    type="messages",
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+# --- Gradio Interface Setup ---
+iface = gr.Interface(
+    fn=generate_response,
+    inputs=[
+        gr.Textbox(lines=4, label="Medical Query (e.g., 'What are the symptoms of type 2 diabetes?')", placeholder="Enter your medical question..."),
+        gr.Slider(minimum=32, maximum=1024, step=32, value=256, label="Max Response Length", info="Controls the length of the generated answer."),
+        gr.Slider(minimum=0.1, maximum=1.0, step=0.1, value=0.7, label="Creativity (Temperature)", info="Higher temperature means more creative/risky answers.")
     ],
+    outputs=gr.Textbox(lines=10, label="Fine-Tuned Medical Assistant Response"),
+    title="⚕️ Medical Llama 2 Fine-Tune Demo (dynamodenis254)",
+    description="This demo uses a Llama 2 model fine-tuned on medical data. Enter a query and observe the specialized response.",
+    theme="soft"
 )
 
-with gr.Blocks() as demo:
-    with gr.Sidebar():
-        gr.LoginButton()
-    chatbot.render()
-
-
+# Launch is handled automatically by Hugging Face Spaces
 if __name__ == "__main__":
-    demo.launch()
+    iface.launch()