from transformers import AutoTokenizer
import transformers
import torch
import streamlit as st
import os

HF_TOKEN = os.environ.get("HF_TOKEN")

prompt = st.text_input('Prompt', 'Hello, How you doing ?')

model = "meta-llama/Llama-2-13b-chat-hf"

#Not Working
#tokenizer = AutoTokenizer.from_pretrained(model)

def load_model(model):
    pipeline = transformers.pipeline(
        "text-generation",
        model=model,
        torch_dtype=torch.float32,
        device_map="auto",
        do_sample=True,
        token=HF_TOKEN,
    )
    return pipeline

def get_llama_response(pipeline,prompt):
    
    sequences = pipeline(
        prompt,
        top_k=10,
        num_return_sequences=1,
        max_length=256,
    )
    print(sequences[0]['generated_text'])
    

pipeline = AutoTokenizer.from_pretrained(model, token=HF_TOKEN)
#prompt="Can you help me to write rest api endpoints in python ?"
response = get_llama_response(pipeline,prompt)

st.write('Answer: ',response)