app.py

import gradio as gr
from huggingface_hub import InferenceClient
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Inicjalizacja InferenceClient
client = InferenceClient("alpindale/WizardLM-2-8x22B")

# Inicjalizacja tokenizera i modelu
model_path = "alpindale/WizardLM-2-8x22B"
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(model_path, device_map="auto").eval()

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    use_local_model: bool,
):
    # Przygotowanie wiadomości do kontekstu
    messages = [{"role": "system", "content": system_message}]
    for user, assistant in history:
        if user:
            messages.append({"role": "user", "content": user})
        if assistant:
            messages.append({"role": "assistant", "content": assistant})
    messages.append({"role": "user", "content": message})

    if use_local_model:
        # Użycie lokalnego modelu
        input_text = "\n".join([f"{m['role']}: {m['content']}" for m in messages])
        input_ids = tokenizer.encode(input_text, return_tensors="pt")
        input_ids = input_ids.to(model.device)
        
        with torch.no_grad():
            output = model.generate(
                input_ids,
                max_new_tokens=max_tokens,
                temperature=temperature,
                top_p=top_p,
                do_sample=True,
                pad_token_id=tokenizer.eos_token_id,
            )
        
        response = tokenizer.decode(output[0], skip_special_tokens=True)
        yield response.split("assistant:")[-1].strip()
    else:
        # Użycie Hugging Face Inference API
        response = ""
        for chunk in client.text_generation(
            "\n".join([f"{m['role']}: {m['content']}" for m in messages]),
            max_new_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
        ):
            response += chunk
            yield response.split("assistant:")[-1].strip()

# Tworzenie interfejsu Gradio
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(
            value="Odpowiadasz w języku polskim. Jesteś Coder/Developer/Programista i tworzysz pełny kod.",
            label="Wiadomość systemowa"
        ),
        gr.Slider(minimum=1, maximum=2048, value=2048, step=1, label="Maksymalna liczba nowych tokenów"),
        gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.1, label="Temperatura"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (próbkowanie nucleus)",
        ),
        gr.Checkbox(label="Użyj lokalnego modelu", value=False),
    ],
    title="Zaawansowany interfejs czatu AI",
    description="Czatuj z modelem AI, korzystając z Hugging Face Inference API lub lokalnego modelu.",
)

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