README.md

---
license: other
pretty_name: 3DA Unified Data Legacy Packed Mirror
tags:
- robotics
- 3da
- legacy
- archive
---

# 3DA Unified Data Legacy Packed Mirror

This dataset repository used to host a packed bootstrap mirror for the 3DA
robotics workspace. The large payloads were intentionally removed to reduce HF
storage usage. The local canonical workspace still contains the source data and
staged tar files needed to recreate the mirror.

Audit / cleanup date: 2026-05-16.

## Removed Payload Layout

```text
SeonghuJeon/3da-unified-data
├── openx_rest_lerobot.tar       # openx_lerobot/lerobot minus droid + _stats + _cache
├── openx_rest_BrunoM42.tar      # openx_lerobot/BrunoM42
├── openx_rest_misc.tar          # openx_lerobot/tailong-wu + HuggingFaceVLA
├── mimicgen.tar                 # data/mimicgen
├── libero_eval.tar              # data/libero eval assets
├── small_files.tar              # checkpoints, model_stats, configs, stats, manifests
├── libero_noop/                 # replayed LIBERO no-op HDF5s
├── _small_files/                # direct convenience copies of selected small files
└── extract.sh
```

## Local Canonical Sources

The payloads can be recreated from the local CVLAB1 workspace:

```text
/NHNHOME/WORKSPACE/0226010404_A/CVLAB/CVLAB1/jisang/
├── data/openx_lerobot/lerobot
├── data/openx_lerobot/_stats
├── data/openx_lerobot/_cache
├── data/openx_lerobot/BrunoM42
├── data/openx_lerobot/tailong-wu
├── data/openx_lerobot/HuggingFaceVLA
├── data/mimicgen
├── data/libero
├── data/libero_noop
├── checkpoints
└── 3DA_unified/
    ├── configs/training
    ├── model_stats
    └── scripts/upload_to_hf.py
```

At cleanup time, these staged files already existed locally:

```text
/NHNHOME/WORKSPACE/0226010404_A/CVLAB/CVLAB1/jisang/3da_hf_staging/
├── openx_rest_lerobot.tar
├── openx_rest_BrunoM42.tar
├── openx_rest_misc.tar
├── mimicgen.tar
├── libero_eval.tar
├── small_files.tar
└── extract.sh
```

## Recreate / Reupload

Dry-run the local structure and estimated sizes:

```bash
cd /NHNHOME/WORKSPACE/0226010404_A/CVLAB/CVLAB1/jisang/3DA_unified
python scripts/upload_to_hf.py --repo-id SeonghuJeon/3da-unified-data --dry-run
```

Rebuild the tar files from local sources and upload them:

```bash
cd /NHNHOME/WORKSPACE/0226010404_A/CVLAB/CVLAB1/jisang/3DA_unified
HF_XET_HIGH_PERFORMANCE=1 python scripts/upload_to_hf.py --force-refresh   --repo-id SeonghuJeon/3da-unified-data
```

If the tar files in `/NHNHOME/.../jisang/3da_hf_staging` are still valid and no
refresh is needed, upload the staging directory directly:

```bash
cd /NHNHOME/WORKSPACE/0226010404_A/CVLAB/CVLAB1/jisang/3DA_unified
HF_XET_HIGH_PERFORMANCE=1 huggingface-cli upload-large-folder   SeonghuJeon/3da-unified-data   /NHNHOME/WORKSPACE/0226010404_A/CVLAB/CVLAB1/jisang/3da_hf_staging   --repo-type dataset
```

## Download If Reuploaded

```bash
python scripts/download_3da_unified_data.py   --repo-id SeonghuJeon/3da-unified-data   --project-root /path/to/project   --data-root /path/to/project/data   --staging /path/to/project/3da_hf_staging
```