Step 6: Upload Dataset to Hugging Face (Optional)

Method 1: Local upload (not recommended, slow upload speed)

• Automatic upload

Set push_to_hub=true when collecting the dataset, and it will be uploaded automatically once collection is complete

• Manual upload

Set push_to_hub=false when collecting the dataset, and upload it manually once collection is complete

hf upload <username>/lerobot_my_dataset_a /Users/<username>/.cache/huggingface/lerobot/<username>/lerobot_my_dataset_a / --repo-type=dataset

Whether automatic or manual, the upload speed is very slow (one hundred KB per second)

Because the HuggingFace servers are overseas

Method 2: Upload via a cloud GPU platform (recommended)

Log in to the cloud GPU platform Featurize

https://featurize.cn?s=d7ce99f842414bfcaea5662a97581bd1

Start a cloud GPU instance

Upload the dataset archive to Datasets

Copy the instance download command

Run in the command line of the cloud GPU instance

pip install httpx

unzip your_datasets.zip

hf auth login

Upload the dataset to HuggingFace

Create an upload_dataset.py file with the following content

from huggingface_hub import HfApi

api = HfApi()

api.upload_folder(
    folder_path="~/lerobot_my_dataset_a",
    repo_id="<username>/lerobot_my_dataset_a",
    repo_type="dataset"
)

api.create_tag("<username>/lerobot_my_dataset_a", tag="v0.4.0", repo_type="dataset")

Run the file

python upload_dataset.py

• Another upload method (not recommended)

hf upload <username>/lerobot_my_dataset_a lerobot_my_dataset_a / --repo-type=dataset

View the dataset on HuggingFace

https://huggingface.co/datasets/Juxi-Technology/soarm_amazing_hand_pick

Related Products

SO-ARM101 Developer Kit→