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

# Model i tokenizer
model_name = "sshleifer/tiny-gpt2"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float32,  # ważne dla CPU
)

# Funkcja do generowania odpowiedzi
def generate_text(prompt):
    inputs = tokenizer(prompt, return_tensors="pt")
    with torch.no_grad():
        output = model.generate(
            **inputs,
            max_new_tokens=150,
            do_sample=True,
            top_p=0.9,
            temperature=0.7,
            pad_token_id=tokenizer.eos_token_id,
        )
    return tokenizer.decode(output[0], skip_special_tokens=True)

# Interfejs Gradio
gr.Interface(
    fn=generate_text,
    inputs=gr.Textbox(lines=5, label="Wprowadź tekst (prompt)"),
    outputs=gr.Textbox(label="Wygenerowany tekst"),
    title="sshleifer/tiny-gpt2 – Generowanie tekstu po polsku",
    description="Testowanie sshleifer/tiny-gpt2 -> Uruchamiany na CPU.",
).launch()