app.py

#!/usr/bin/env python3
"""
HuggingFace Space Gradio应用
RVC v2 训练界面
"""

import os
import gradio as gr
import subprocess
import threading
from pathlib import Path

# 创建必要的目录
Path("checkpoints").mkdir(exist_ok=True)
Path("data/training_data").mkdir(parents=True, exist_ok=True)
Path("logs").mkdir(exist_ok=True)

# 训练状态
training_status = "未开始"
training_log = []

def get_training_status():
    """获取训练状态"""
    checkpoints_dir = Path("checkpoints")
    logs_dir = Path("logs")

    status_lines = []
    status_lines.append(f"## 当前状态: {training_status}")

    # 检查checkpoints
    if checkpoints_dir.exists():
        checkpoints = list(checkpoints_dir.glob("*.pth"))
        if checkpoints:
            status_lines.append(f"### 模型文件: {len(checkpoints)} 个")
            for ckpt in checkpoints:
                size_mb = ckpt.stat().st_size / 1024 / 1024
                status_lines.append(f"- {ckpt.name}: {size_mb:.2f} MB")
        else:
            status_lines.append("### 模型文件: 尚未生成")
    else:
        status_lines.append("### 模型文件: 目录不存在")

    # 检查日志
    if logs_dir.exists():
        log_files = list(logs_dir.glob("*.txt"))
        if log_files:
            status_lines.append(f"### 日志文件: {len(log_files)} 个")
            latest_log = max(log_files, key=lambda p: p.stat().st_mtime)
            status_lines.append(f"最新日志: {latest_log.name}")
        else:
            status_lines.append("### 日志文件: 尚未生成")
    else:
        status_lines.append("### 日志文件: 目录不存在")

    # 检查数据
    data_dir = Path("data/training_data/audio")
    if data_dir.exists():
        audio_files = list(data_dir.glob("*.wav")) + list(data_dir.glob("*.mp3"))
        status_lines.append(f"### 训练数据: {len(audio_files)} 个音频文件")
    else:
        status_lines.append("### 训练数据: 尚未下载")

    return "\n".join(status_lines)

def start_training():
    """启动训练"""
    global training_status

    if training_status == "正在训练":
        return "⚠️ 训练已在进行中"

    training_status = "正在训练"

    # 在后台线程中运行训练
    def train_thread():
        global training_status
        try:
            # 运行训练脚本
            result = subprocess.run(
                ["python", "scripts/train_rvc_v2_fixed.py"],
                capture_output=True,
                text=True,
                timeout=14400  # 4小时超时
            )

            # 保存日志
            with open("logs/training_log.txt", "w", encoding="utf-8") as f:
                f.write("=== STDOUT ===\n")
                f.write(result.stdout)
                f.write("\n=== STDERR ===\n")
                f.write(result.stderr)

            if result.returncode == 0:
                training_status = "训练完成"
            else:
                training_status = f"训练失败: {result.returncode}"

        except Exception as e:
            training_status = f"训练异常: {str(e)}"

    thread = threading.Thread(target=train_thread)
    thread.daemon = True
    thread.start()

    return "✅ 训练已启动"

def download_data():
    """下载训练数据"""
    try:
        result = subprocess.run(
            ["python", "scripts/prepare_training_data_v2.py"],
            capture_output=True,
            text=True,
            timeout=600  # 10分钟超时
        )

        if result.returncode == 0:
            return "✅ 数据下载完成\n\n" + result.stdout
        else:
            return "❌ 数据下载失败\n\n" + result.stderr

    except Exception as e:
        return f"❌ 数据下载异常: {str(e)}"

# 创建Gradio界面
with gr.Blocks(title="NumberBlocks One RVC v2 训练") as demo:
    gr.Markdown("# 🎤 NumberBlocks One RVC v2 训练")
    gr.Markdown("## 使用9小时One语音数据训练RVC v2模型")

    with gr.Tab("训练状态"):
        status_output = gr.Markdown(value=get_training_status())
        refresh_btn = gr.Button("刷新状态")
        refresh_btn.click(get_training_status, outputs=status_output)

    with gr.Tab("开始训练"):
        gr.Markdown("### 步骤1: 下载训练数据")
        download_btn = gr.Button("下载数据")
        download_output = gr.Textbox(label="下载结果", lines=10)
        download_btn.click(download_data, outputs=download_output)

        gr.Markdown("### 步骤2: 开始训练")
        train_btn = gr.Button("开始训练", variant="primary")
        train_output = gr.Textbox(label="训练状态")
        train_btn.click(start_training, outputs=train_output)

    with gr.Tab("训练配置"):
        gr.Markdown("### 训练参数")
        gr.Markdown("""
        - **模型**: RVC v2
        - **数据集**: ayf3/numberblocks-audio (9小时音频)
        - **硬件**: T4 GPU (免费)
        - **训练时长**: 2-3小时
        - **Batch Size**: 8
        - **Epochs**: 500
        - **学习率**: 0.0001
        - **优化器**: AdamW
        """)

# 启动Gradio应用
if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)