upload_to_hf.py

#!/usr/bin/env python3
"""
上传 LeRobot 数据集到 Hugging Face Hub
"""

import os
from huggingface_hub import HfApi, create_repo
from pathlib import Path

def upload_dataset_to_hf(
    dataset_path: str,
    repo_name: str,
    username: str = None,
    private: bool = False
):
    """
    上传数据集到 Hugging Face Hub
    
    Args:
        dataset_path: 数据集本地路径
        repo_name: 仓库名称
        username: Hugging Face 用户名 (可选)
        private: 是否设为私有仓库
    """
    
    # 初始化 HF API
    api = HfApi()
    
    # 构建完整的仓库名称
    if username:
        full_repo_name = f"{username}/{repo_name}"
    else:
        # 获取当前用户信息
        user_info = api.whoami()
        username = user_info["name"]
        full_repo_name = f"{username}/{repo_name}"
    
    print(f"准备上传到: {full_repo_name}")
    
    # 创建仓库 (如果不存在)
    try:
        create_repo(
            repo_id=full_repo_name,
            repo_type="dataset",
            private=private,
            exist_ok=True
        )
        print(f"✅ 仓库 {full_repo_name} 已创建或已存在")
    except Exception as e:
        print(f"⚠️ 创建仓库时出现问题: {e}")
    
    # 上传所有文件
    dataset_path = Path(dataset_path)
    
    # 获取所有需要上传的文件
    files_to_upload = []
    
    # 添加 meta 目录下的文件
    meta_dir = dataset_path / "meta"
    if meta_dir.exists():
        for file_path in meta_dir.iterdir():
            if file_path.is_file():
                files_to_upload.append(str(file_path))
    
    # 添加 data 目录下的所有 parquet 文件
    data_dir = dataset_path / "data"
    if data_dir.exists():
        for chunk_dir in data_dir.iterdir():
            if chunk_dir.is_dir():
                for file_path in chunk_dir.iterdir():
                    if file_path.is_file() and file_path.suffix == ".parquet":
                        files_to_upload.append(str(file_path))
    
    # 添加 README.md
    readme_path = dataset_path / "README.md"
    if readme_path.exists():
        files_to_upload.append(str(readme_path))
    
    print(f"📁 找到 {len(files_to_upload)} 个文件需要上传")
    
    # 上传文件
    try:
        api.upload_folder(
            folder_path=dataset_path,
            repo_id=full_repo_name,
            repo_type="dataset",
            path_in_repo="",
            commit_message="Upload LeRobot dataset"
        )
        print(f"✅ 数据集已成功上传到 {full_repo_name}")
        print(f"🔗 访问链接: https://huggingface.co/datasets/{full_repo_name}")
        
    except Exception as e:
        print(f"❌ 上传失败: {e}")
        return False
    
    return True

if __name__ == "__main__":
    # 配置参数
    DATASET_PATH = "/home/shuo/research/datasets/steer_test_lerobot"
    REPO_NAME = "steer_test_lerobot"  # 修改为您想要的仓库名
    USERNAME = "TreeePlanter"  # 设置为您的 HF 用户名，或留空自动检测
    PRIVATE = False  # 设为 True 创建私有仓库
    
    # 检查数据集路径
    if not os.path.exists(DATASET_PATH):
        print(f"❌ 数据集路径不存在: {DATASET_PATH}")
        exit(1)
    
    print("🚀 开始上传数据集到 Hugging Face Hub...")
    print(f"📂 数据集路径: {DATASET_PATH}")
    print(f"📦 仓库名称: {REPO_NAME}")
    
    success = upload_dataset_to_hf(
        dataset_path=DATASET_PATH,
        repo_name=REPO_NAME,
        username=USERNAME,
        private=PRIVATE
    )
    
    if success:
        print("\n🎉 上传完成！")
        print("\n📋 后续步骤:")
        print("1. 访问您的数据集页面")
        print("2. 检查数据集是否正确显示")
        print("3. 添加标签和描述")
        print("4. 测试数据集加载:")
        print("   from datasets import load_dataset")
        print(f"   dataset = load_dataset('{USERNAME or 'your-username'}/{REPO_NAME}')")
    else:
        print("\n❌ 上传失败，请检查错误信息")