app.py

import os
import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
from huggingface_hub import login

login(token=os.getenv('HF_TOKEN'))

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("Zyphra/Zamba2-7B")
model = AutoModelForCausalLM.from_pretrained(
    "Zyphra/Zamba2-7B",
    device_map="auto",  # Automatically handles device placement
    torch_dtype=torch.bfloat16
)

def generate_response(input_text):
    input_ids = tokenizer(input_text, return_tensors="pt").to(model.device)
    outputs = model.generate(
        **input_ids,
        max_new_tokens=500,
        do_sample=True,
        temperature=0.7,
        top_k=50,
        top_p=0.9,
        repetition_penalty=1.2,
        num_beams=5,
        length_penalty=1.0,
        num_return_sequences=1
    )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response

# Create the Gradio interface
demo = gr.Interface(
    fn=generate_response,
    inputs=gr.inputs.Textbox(lines=5, placeholder="Enter your question here..."),
    outputs="text",
    title="Zamba2-7B Model",
    description="Ask Zamba2 7B a question."
)

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