app.py

import gradio as gr
import spaces
import torch

import transformers
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "microsoft/Phi-3-mini-128k-instruct"

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, trust_remote_code=True)

pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    device="cpu",
    trust_remote_code=True
)

@spaces.GPU
def chat_function(message, history, system_prompt,max_new_tokens,temperature):
    prompt = system_prompt
    for msg in history:
        prompt += f'{msg["role"]}: {msg["content"]}\n'
    prompt += f'user: {message}\n'
    
    temp = temperature + 0.1
    outputs = pipeline(
        prompt,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        temperature=temp,
        top_p=0.9,
    )
    generated_text = outputs[0]['generated_text']
    new_text = generated_text[len(prompt):]
    return new_text

gr.ChatInterface(
    chat_function,
    chatbot=gr.Chatbot(height=400),
    textbox=gr.Textbox(placeholder="Enter message here", container=False, scale=7),
    title="microsoft/Phi-3-mini-128k-instruct",
    description="""
    This space is dedicated for chatting with Meta's Latest LLM - Llama 8b Instruct. Find this model here: https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct 
    Feel free to play with customization in the "Additional Inputs".
    """,
    theme="soft",
    additional_inputs=[
        gr.Textbox("You are helpful AI.", label="System Prompt"),
        gr.Slider(512, 4096, label="Max New Tokens"),
        gr.Slider(0, 1, label="Temperature")
    ]
).launch()