import gradio as gr
import torch
from peft import PeftModel, PeftConfig
from transformers import AutoTokenizer, AutoModelForCausalLM


#Loading model
device = "cuda" if torch.cuda.is_available() else "cpu"
model_path = "parsanna17/finetune_starcoder2_with_R_data"
checkpoint = "bigcode/starcoder2-3b"
config = PeftConfig.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(checkpoint,device=device, torch_dtype=torch.bfloat16)
model = PeftModel.from_pretrained(model, model_path).to(device)
tokenizer = AutoTokenizer.from_pretrained(checkpoint)

if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

def remove_header_trailer(input):
  text = input.split()
  start=0
  end=0
  i=0
  while i<len(text)-1 and text[i]!="#Solution:" : 
      i+=1
  start =i+1
  i+=1
  while i<len(text)-1 and text[i]!="Solution:" and text[i]!="#Question:" and text[i]!=text[i+1] : 
      i+=1
  end = i+1 if len(text)==i else i
  text= text[start:end]
  return " ".join(text)

def generate(inputs):
    prompt = f"""Write a code as R programmer.
    #Context: You are a R Programmer going for an interview you need to provide code snippet for the given question in R programming Language.
    #Question: create a function to {inputs} in R language
    #Solution: """
    inputs = tokenizer(prompt, return_tensors="pt").to(device)
    with torch.no_grad():
      outputs = model.generate(**inputs, pad_token_id=tokenizer.pad_token_id,max_new_tokens=100)
    return remove_header_trailer(tokenizer.decode(outputs[0]))

demo = gr.Interface(fn = generate, 
                    inputs = gr.Textbox(lines=5, placeholder = "write you program details to generate code in R", label="Code Prompt"),
                    outputs=gr.Textbox(lines=5,placeholder = "Code will be generated here", label="R Code"),
                    title="R Programming Language Code Generator",
    description="Code is being generated using Starcoder2-3b llm fine tuned on Kaggle using R dataset",
    article = "Created and Maintained By Prasanna Dhungana")

demo.launch()