app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

# Use a CPU-compatible base model (replace this with your actual full-precision model)
base_model_id = "unsloth/gemma-2-9b"  # Replace with real CPU-compatible model
lora_model_id = "import gradio as gr"
from huggingface_hub import InferenceClient
import os

# 🔹 Hugging Face Credentials
HF_REPO = "Futuresony/gemma2-9b-lora-alpaca"
HF_TOKEN = os.getenv('HUGGINGFACEHUB_API_TOKEN')

client = InferenceClient(HF_REPO, token=HF_TOKEN)

def format_alpaca_prompt(user_input, system_prompt, history):
    """Formats input in Alpaca/LLaMA style"""
    history_str = "\n".join([f"### Instruction:\n{h[0]}\n### Response:\n{h[1]}" for h in history])
    prompt = f"""{system_prompt}
{history_str}

### Instruction:
{user_input}

### Response:
"""
    return prompt

def respond(message, history, system_message, max_tokens, temperature, top_p):
    formatted_prompt = format_alpaca_prompt(message, system_message, history)

    response = client.text_generation(
        formatted_prompt,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
    )

    # ✅ Extract only the response
    cleaned_response = response.split("### Response:")[-1].strip()
    
    history.append((message, cleaned_response))  # ✅ Update history with the new message and response
    
    yield cleaned_response  # ✅ Output only the answer

demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=250, value=128, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.9, step=0.1, label="Temperature"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.99, step=0.01, label="Top-p (nucleus sampling)"),
    ],
)

if __name__ == "__main__":
    demo.launch()"

# Load the base model on CPU
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_id,
    torch_dtype=torch.float32,  # Use float32 for CPU compatibility
    device_map="cpu"
)

# Load the PEFT LoRA model
model = PeftModel.from_pretrained(base_model, lora_model_id)

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(base_model_id)

# Chat function
def respond(message, history: list[tuple[str, str]], system_message, max_tokens, temperature, top_p):
    messages = [{"role": "system", "content": system_message}]
    for user_msg, bot_msg in history:
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if bot_msg:
            messages.append({"role": "assistant", "content": bot_msg})
    messages.append({"role": "user", "content": message})

    # Generate response (simulated loop for streaming output)
    inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cpu")
    outputs = model.generate(
        inputs,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
    )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    yield response

# Gradio UI
demo = gr.ChatInterface(
    fn=respond,
    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"),
    ],
)

if __name__ == "__main__":
    demo.launch()