import gradio as gr
from transformers import pipeline
import torch
import os

token = os.environ.get('HF_TOKEN')

print('🔄 Loading Gemma 31B...')
pipe = pipeline(
    'text-generation',
    model='Arabic250/gemma4',
    torch_dtype=torch.bfloat16,
    device_map='auto',
    token=token
)

def predict(text):
    if not text: return "يرجى إدخال نص"
    res = pipe(text, max_new_tokens=256, do_sample=True, temperature=0.7)
    return res[0]['generated_text']

demo = gr.Interface(fn=predict, inputs="text", outputs="text", title="Gemma 4 31B Studio")
demo.launch()