import streamlit as st

import torch
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "NL2SQL_BLOOMZ-3B"
HUGGING_FACE_USER_NAME = "abhishek23HF"

peft_model_id = f"{HUGGING_FACE_USER_NAME}/{model_name}"
config = PeftConfig.from_pretrained(peft_model_id)
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, return_dict=True, load_in_8bit=False)
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)

# Load the Lora model
model = PeftModel.from_pretrained(model, peft_model_id)


# from IPython.display import display, Markdown

def make_inference(db_id, question):
  batch = tokenizer(f"""
  ### INSTRUCTION\n
  Below is a User Question for a SQL DATABASE. Your job is to write a SQL Query for the given question from the user for that particular Database.
  \n\n
  ### DATABASE_ID:\n{db_id}\n
  ### USER QUESTION:\n{question}\n\n
  ### SQL QUERY:\n
  """, return_tensors='pt')

  with torch.cuda.amp.autocast():
      output_tokens = model.generate(**batch, max_new_tokens=200)

  return tokenizer.decode(output_tokens[0], skip_special_tokens=True)



# Create two text input boxes
text_input_db_id= st.text_input("DB ID")
text_input_question = st.text_input("User Query")

# make_inference(your_db_id_here, your_db_query_here)

# Display the text input boxes

if st.button('Submit'):
    st.write(make_inference(text_input_db_id, text_input_question))