app.py

import streamlit as st
from transformers import AutoModelForCausalLM, AutoTokenizer
import accelerate

# Load the model and tokenizer
@st.cache_resource
def load_model_and_tokenizer():
    model_name_or_path = "anthropic/mistral-7b"
    accelerator = accelerate.Accelerator(device_map="auto")
    model = AutoModelForCausalLM.from_pretrained(model_name_or_path, device_map=accelerator.device_map)
    tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
    return model, tokenizer

# Function to generate the response
@st.cache_data
def generate_response(prompt):
    prompt_template = f'''
    <|prompter|>:{prompt}
    <|assistant|>:
    '''
    input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids
    with accelerator.autocast():
        output = model.generate(inputs=input_ids, temperature=0.7, do_sample=True, eos_token_id=tokenizer.eos_token_id, pad_token_id=tokenizer.pad_token_id, max_new_tokens=512)
    response = tokenizer.decode(output[0], skip_special_tokens=True)
    return response

# Streamlit app
def main():
    st.title("Mistral 7B Language Model")
    model, tokenizer = load_model_and_tokenizer()

    prompt = st.text_area("Enter your query:")
    if st.button("Submit"):
        with st.spinner("Generating response..."):
            response = generate_response(prompt)
            st.write(response)

if __name__ == "__main__":
    main()