app.py

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

model_name = "mradermacher/YandexGPT-5-Lite-8B-instruct-GGUF"
filename = "YandexGPT-5-Lite-8B-instruct.Q8_0.gguf"
torch_dtype = torch.float32
tokenizer = AutoTokenizer.from_pretrained(model_name, gguf_file=filename)
model = AutoModelForCausalLM.from_pretrained(model_name, gguf_file=filename, torch_dtype=torch_dtype)

def generate_text(input_text):
    inputs = tokenizer(input_text, return_tensors="pt")
    outputs = model.generate(
        **inputs,
        max_new_tokens=300,  
        do_sample=True,      
        temperature=0.7      
    )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

interface = gr.Interface(
    fn=generate_text,
    inputs=gr.Textbox(lines=2, placeholder="Введите ваш запрос..."),
    outputs="text",
    title="Saiga YandexGPT 8B Demo",
    description="Задайте вопрос модели Saiga YandexGPT 8B!"
)

interface.launch()