app.py

# Подключение клиентов
# - - - - - - - - - - - - - -
from huggingface_hub import InferenceClient
from together import Together

# Подключение библиотек
# - - - - - - - - - - - - - -
import requests
import gradio as gr
import os
import json


#============================
#============================


# Список доступных моделей
# - - - - - - - - - - - - - -
models = {
    "together": [
        "deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free",
        "meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
    ],
    "huggingface": [
        "google/gemma-3-27b-it",
        "Qwen/QwQ-32B",
        "Qwen/QwQ-32B-Preview",
        "mistralai/Mistral-Small-24B-Instruct-2501",
        "deepseek-ai/deepseek-llm-67b-chat",
        "mistralai/Mixtral-8x22B-Instruct-v0.1",
        "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO",
        "google/gemma-2-9b-it",
        "google/gemma-2-27b-it"
    ]
}


#============================
#============================


# Функции для работы с сообщениями
# - - - - - - - - - - - - - -
def add_message(role, content, messages):
    messages.append({"role": role, "content": content})
    return messages, len(messages), str(messages)

def clear_messages(messages):
    return [], 0, "[]"

def show_messages(messages):
    return str(messages)

def get_messages_api(messages):
    return json.dumps(messages, indent=4)

def run_huggingface_model(model, messages, max_tokens, temperature, top_p):
    API_TOKEN = os.getenv("HF_READ_TOKEN")
    headers = {"Authorization": f"Bearer {API_TOKEN}"}
    
    payload = {
        "messages": messages,
        "max_tokens": max_tokens,
        "temperature": temperature,
        "top_p": top_p,
        "stream": False
    #    "inputs": json.dumps(messages),
    #    "seed": random.randint(1, 1000000000),
    #    "parameters": {
    #        "max_tokens": max_tokens,
    #        "temperature": temperature,
    #        "top_p": top_p
    #    }
    }

    model = "https://api-inference.huggingface.co/models/" + model + "/v1/chat/completions";
    response = requests.post(model, headers=headers, json=payload, timeout=30)

    if response.status_code != 200:
        print("ERROR: " + response["error"])
    else:
        print(response.content)
        response = json.loads(response.content)
        result = response["choices"][0]["message"]["content"]
        return result

def run_together_model(model, messages, max_tokens, temperature, top_p):
    client = Together()
    response = client.chat.completions.create(
        model=model,
        messages=messages,
        max_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
    )
    return response.choices[0].message.content


#============================
#============================


# Создаем интерфейс с вкладками
demo = gr.Blocks()

with demo:
    gr.Markdown("# Chat Interface")
    
    # Вкладки для Together и HuggingFace
    with gr.Tabs():
        with gr.Tab("Together"):
            together_model_input = gr.Radio(
                label="Select a Together model",
                choices=models["together"],
                value=models["together"][0],
            )
            together_run_button = gr.Button("Run Together")
        
        with gr.Tab("HuggingFace"):
            huggingface_model_input = gr.Radio(
                label="Select a HuggingFace model",
                choices=models["huggingface"],
                value=models["huggingface"][0],
            )
            huggingface_run_button = gr.Button("Run HuggingFace")
    
    # Общие элементы интерфейса
    role_input = gr.Dropdown(
        label="Role",
        choices=["system", "user", "assistant"],  # Список ролей
        value="user"  # Значение по умолчанию
    )
    content_input = gr.Textbox(label="Content")
    messages_state = gr.State(value=[])
    messages_output = gr.Textbox(label="Messages", value="[]")
    count_output = gr.Number(label="Count", value=0)
    response_output = gr.Textbox(label="Response")
    messages_api_output = gr.Textbox(label="Messages API")

    add_button = gr.Button("Add")
    clear_button = gr.Button("Clear")
    show_button = gr.Button("Show messages")
    get_api_button = gr.Button("Get messages API")

    max_tokens_slider = gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens")
    temperature_slider = gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature")
    top_p_slider = gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")

    # Обработчики событий для кнопок
    add_button.click(
        add_message,
        inputs=[role_input, content_input, messages_state],
        outputs=[messages_state, count_output, messages_output],
    )

    clear_button.click(
        clear_messages,
        inputs=[messages_state],
        outputs=[messages_state, count_output, messages_output],
    )

    show_button.click(
        show_messages,
        inputs=[messages_state],
        outputs=[messages_output],
    )

    get_api_button.click(
        get_messages_api,
        inputs=[messages_state],
        outputs=[messages_api_output],
    )

    # Обработчики событий для кнопок "Run"
    together_run_button.click(
        run_together_model,
        inputs=[together_model_input, messages_state, max_tokens_slider, temperature_slider, top_p_slider],
        outputs=[response_output],
    )

    huggingface_run_button.click(
        run_huggingface_model,
        inputs=[huggingface_model_input, messages_state, max_tokens_slider, temperature_slider, top_p_slider],
        outputs=[response_output],
    )


#============================
#============================


if __name__ == "__main__":
    demo.launch()