app.py

import gradio as gr
import requests
import os

HF_URL = "http://localhost:7860/completion"  # llama-server inside container

def translate(text, direction):
    if direction == "English → Japanese":
        prompt = f"Translate to Japanese:\n{text}\nJapanese:"
    else:
        prompt = f"Translate to English:\n{text}\nEnglish:"

    resp = requests.post(
        HF_URL,
        json={
            "prompt": prompt,
            "n_predict": 128,
            "temperature": 0.1,
            "top_p": 0.9,
            "repeat_penalty": 1.3,
            "stop": ["\n\n","</s>"]
        }
    )
    return resp.json().get("content", "")

with gr.Blocks() as demo:
    gr.Markdown("# 🇬🇧⇄🇯🇵 English ↔ Japanese Translator (llama.cpp)")
    with gr.Row():
        with gr.Column():
            direction = gr.Radio(["English → Japanese", "Japanese → English"], value="English → Japanese")
            input_box = gr.Textbox(label="Input")
            output_box = gr.Textbox(label="Translation")
            run_btn = gr.Button("Translate")
    run_btn.click(fn=translate, inputs=[input_box, direction], outputs=output_box)

demo.launch(server_name="0.0.0.0", server_port=7860)