app.py

import gradio as gr
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "thrishala/mental_health_chatbot"

try:
    model = AutoModelForCausalLM.from_pretrained(
        model_id,
        device_map="cpu",
        torch_dtype=torch.float16,
        low_cpu_mem_usage=True,
        max_memory={"cpu": "15GB"},
        offload_folder="offload",
    )
    
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    tokenizer.model_max_length = 256  # Set maximum length
    
    pipe = pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        torch_dtype=torch.float16,
        num_return_sequences=1,
        do_sample=False,
        truncation=True,
        max_new_tokens=128
    )

except Exception as e:
    print(f"Error loading model: {e}")
    exit()

def respond(message, history, system_message, max_tokens, temperature, top_p):
    prompt = f"{system_message}\n"
    for user_msg, bot_msg in history:
        prompt += f"User: {user_msg}\nAssistant: {bot_msg}\n"
    prompt += f"User: {message}\nAssistant:"

    try:
        response = pipe(
            prompt,
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            do_sample=False,
            pad_token_id=tokenizer.eos_token_id
        )[0]["generated_text"]
        
        bot_response = response.split("Assistant:")[-1].strip()
        yield bot_response
    except Exception as e:
        print(f"Error during generation: {e}")
        yield "An error occurred during generation."

demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(
            value="You are a friendly and helpful mental health chatbot.",
            label="System message",
        ),
        gr.Slider(minimum=1, maximum=128, value=128, 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)",
        ),
    ],
    chatbot=gr.Chatbot(type="messages"),  # Updated to new format
)

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