lehome_submission

LeHome Challenge Submission

This repository contains a Hugging Face submission package for the LeHome Challenge.

Contact

• Contact email: [email protected]

Included Contents

• checkpoints/pretrained_model/ - exported policy checkpoint
• scripts/ - evaluation entrypoint and policy loading code
• configs/train_smolvla_continue_top_short_3k.yaml - training configuration used for this checkpoint
• pyproject.toml - Python dependencies for the project root
• source/lehome/ - local LeHome package required by the evaluation code
• Datasets/example/top_short_merged/meta/ - dataset metadata required for LeRobot policy evaluation

Policy Information

• Policy type: lerobot
• Checkpoint path: checkpoints/pretrained_model
• Dataset metadata path: Datasets/example/top_short_merged/meta
• Recommended dataset root for evaluation: Datasets/example/top_short_merged

Setup

1. Clone or download this repository

git clone https://huggingface.co/<your_username>/<your_repo_name>
cd <your_repo_name>

2. Create environment and install dependencies

Use the same environment setup as the main project. At minimum, install the dependencies declared in pyproject.toml and make sure source/lehome is available on PYTHONPATH or installed as a package.

Example:

pip install -e source/lehome
pip install -e .

If your evaluation environment follows the official LeHome setup, use that environment and then place this repository contents inside the workspace.

Evaluation

Example command:

python -m scripts.eval \
    --policy_type lerobot \
    --policy_path checkpoints/pretrained_model \
    --garment_type top_short \
    --dataset_root Datasets/example/top_short_merged \
    --num_episodes 2 \
    --enable_cameras \
    --device cpu

Notes

• --policy_path must point to the pretrained_model directory, not just to model.safetensors.
• Keep the Datasets/example/top_short_merged/meta directory even if the full training dataset is removed.
• If additional private assets or credentials are needed, provide them in this README.

Upload to Hugging Face

From the parent workspace containing this folder:

hf auth login
hf repo create <your_username>/<your_repo_name> --type model
hf upload-large-folder <your_username>/<your_repo_name> submission_hf --repo-type model