app.py

import gradio as gr
from huggingface_hub import InferenceClient
import os
"""
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
"""
import requests

from openai import OpenAI, AsyncOpenAI

clients = {}

try:
    clients['32B-Pro (beta)'] = [OpenAI(api_key='123', base_url=os.getenv('MODEL_NAME_OR_PATH_32B')), requests.get(os.getenv('MODEL_NAME_OR_PATH_32B') + '/models').json()['data'][0]['id']]
except:
    pass

try:
    clients['32B QWQ (experimental, without any additional tuning after LEP!)'] = [OpenAI(api_key='123', base_url=os.getenv('MODEL_NAME_OR_PATH_QWQ')), requests.get(os.getenv('MODEL_NAME_OR_PATH_QWQ') + '/models').json()['data'][0]['id']]
except:
    pass
    
try:
    clients['7B (work in progress)'] = [OpenAI(api_key='123', base_url=os.getenv('MODEL_NAME_OR_PATH_7B')), requests.get(os.getenv('MODEL_NAME_OR_PATH_7B') + '/models').json()['data'][0]['id']]
except:
    pass
    
try:
    clients['3B'] = [OpenAI(api_key='123', base_url=os.getenv('MODEL_NAME_OR_PATH_3B')), requests.get(os.getenv('MODEL_NAME_OR_PATH_3B') + '/models').json()['data'][0]['id']]
except:
    pass

def respond(
    message,
    history: list[tuple[str, str]],
    model_name,
    system_message,
    max_tokens,
    temperature,
    top_p,
    repetition_penalty
):
    messages = []
    if len(system_message.strip()) > 0:
        messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

    messages.append({"role": "user", "content": message})

    response = ""

    res = clients[model_name][0].chat.completions.create(
        model=clients[model_name][1],
        messages=messages,
        temperature=temperature,
        top_p=top_p,
        max_tokens=max_tokens,
        stream=True,
        extra_body={
            "repetition_penalty": repetition_penalty,
            "add_generation_prompt": True,
        }
    )

    for message in res:
        token = message.choices[0].delta.content

        response += token
        yield response


"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
options = ["32B-Pro (beta)", "32B QWQ (experimental, without any additional tuning after LEP!)", "7B (work in progress)", "3B"]
options = options[:2]
system_old = "You are a helpful and harmless assistant. You should think step-by-step. First, reason (the user does not see your reasoning), then give your final answer."
system_new = "Ты Руадапт - полезный и дружелюбный интеллектуальный ассистент для помощи пользователям в их вопросах."
system_new2 = "Ты — Руадапт, русскоязычный автоматический ассистент. Ты разговариваешь с людьми и помогаешь им."
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Radio(choices=options, label="Model:", value=options[0]),
        gr.Textbox(value=system_new2, label="System message"),
        gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.0, maximum=2.0, value=0.3, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
        gr.Slider(minimum=0.9, maximum=1.5, value=1.05, step=0.05, label="repetition_penalty"),
    ],
    concurrency_limit=10
)


if __name__ == "__main__":
    #print(requests.get(os.getenv('MODEL_NAME_OR_PATH')[:-3] + '/docs'))
    demo.launch(share=True)