app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
import torch
from peft import PeftModel, PeftConfig, LoraConfig  # Import PeftModel only once

# Model paths
base_model_name = "microsoft/phi-2"
new_model_path = "Sunirmala/phi2-Phi2Med"  # Adjusted path for clarity

# Load tokenizer and model
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained( "microsoft/phi-2", trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"

from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM

config = PeftConfig.from_pretrained("Sunirmala/phi2-Phi2Med")
model1 = AutoModelForCausalLM.from_pretrained(base_model_name, trust_remote_code=True)
model = PeftModel.from_pretrained(model1, "Sunirmala/phi2-Phi2Med")

# Define the chat function using appropriate prompt format
def QLoRA_Chatgpt(prompt):
    print(prompt)
    pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer, max_length=200)
    result = pipe(f"<s>[INST] {prompt} [/INST]")
    return(result[0]['generated_text'])
    

# Define Interface
description = 'An AI assistant that works on the Microsoft Phi 2 model, which has been finetuned on the MedQuad dataset using the QLora method, operates effectively. Created by Sunirmala Mohanta and Junaid Naeemulla Khan'
title = 'AI Chat bot finetuned on Microsoft Phi 2 model using QLORA'
iface = gr.Interface(
                    fn=QLoRA_Chatgpt, 
                    inputs=gr.Textbox("how can help you today", label='prompt'), 
                    outputs=gr.Textbox(label='Generated-output',scale = 2), 
                    title = title,
                    description = description)
iface.launch(share=True)