app.py

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, pipeline
import gradio as gr
import torch

model = AutoModelForCausalLM.from_pretrained(
    "microsoft/Phi-3.5-mini-instruct",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True
)
model.load_adapter('./phi3_custom_oasst1_v1')

tokenizer = AutoTokenizer.from_pretrained('./phi3_custom_oasst1_v1', trust_remote_code=True)

def generateText(inputText, num_tokens=200):

    pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer, max_length=num_tokens)
    result = pipe(f'''{inputText}''')
    return result[0]['generated_text']
    

title = "Fine tuned Phi3.5 instruct model on OpenAssist dataset using QLora"
description = '''
Phi-3.5-mini is a lightweight, state-of-the-art open model built upon datasets used for Phi-3 - synthetic data and filtered publicly available websites - with a focus on very high-quality, reasoning dense data. 
The model belongs to the Phi-3 model family and supports 128K token context length. 
The model underwent a rigorous enhancement process, incorporating both supervised fine-tuning, proximal policy optimization, and direct preference optimization to ensure precise instruction adherence and robust safety measures.
This demo utilises a fine tuned version of Phi3.5 instruct model using QLora on OpenAssist dataset - a human-generated, human-annotated assistant-style conversation corpus consisting of 161,443 messages in 35 different languages, annotated with 461,292 quality ratings, resulting in over 10,000 fully annotated conversation trees.
'''
examples = [
            ["How do I build a PC?", 200],
            ["write me a top 10 list of the funniest ways to die?", 300],
            ["Write a response that disagrees with the following post: 'Technology is everything that doesn't work yet.'?", 300]
           ]

demo = gr.Interface(
    generateText, 
    inputs = [
        gr.Textbox(label="User Question"),
        gr.Slider(100, 500, value = 200, step=100, label="Number of Output tokens"),
        ], 
    outputs = [
        gr.Text(),
        ],
    title = title,
    description = description,
    examples = examples
)
demo.launch()