Instructions to use AlexWortega/tinyvla with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use AlexWortega/tinyvla with LeRobot:
- Notebooks
- Google Colab
- Kaggle
wds mixture: unitree+nav converters (gnm/vamos), wds source branch in train_fast, AMP autoselect (V100 fp16), embedding expand 16->32, push/fetch/validate tooling, DATA.md
b152c62 verified | # Параллельная unitree-фаза: build_shards.py (v3) однопоточный по своей природе, | |
| # так что параллелим ПО СЕТАМ (xargs -P). Каждый сет: build -> pack -> validate, | |
| # tmp-шарды удаляются сразу. Дисковый предохранитель: ждать, пока свободно <30GB. | |
| set -uo pipefail | |
| cd /mnt/storage/tinyvla_wds | |
| export HF_HOME=/mnt/storage/tinyvla_wds/.hf_home | |
| source venv/bin/activate | |
| S=tinyvla_h100/scripts | |
| mkdir -p data/wds/unitree data/shards_tmp logs | |
| UNITREE_G1="G1_Dex1_StackRygBlock_Dataset_Sim G1_Dex1_DiverseManip_DualArm_256x256 \ | |
| G1_Dex1_MountCamera_Dataset G1_Dex3_ToastedBread_Dataset G1_Dex1_DiverseManip_SingleArm_256x256 \ | |
| G1_Brainco_PickDoll_Dataset G1_Dex3_PickDoll_Dataset G1_Dex3_BlockStacking_Dataset \ | |
| G1_Dex3_GraspSquare_Dataset G1_Dex3_CameraPackaging_Dataset G1_Brainco_GraspOreo_Dataset" | |
| UNITREE_Z1="Z1_StackBox_Dataset Z1_Dual_Dex1_PourCoffee_Dataset Z1_Dual_Dex1_FoldClothes_Dataset" | |
| one() { | |
| name=$1 | |
| [ -f data/wds/unitree/$name/manifest.json ] && { echo "skip $name (done)"; return 0; } | |
| case " Z1_StackBox_Dataset Z1_Dual_Dex1_PourCoffee_Dataset Z1_Dual_Dex1_FoldClothes_Dataset " in | |
| *" $name "*) EMB=19;; *) EMB=15;; | |
| esac | |
| while [ "$(df --output=avail -BG /mnt | tail -1 | tr -dc 0-9)" -lt 30 ]; do | |
| echo "$name: disk low, waiting"; sleep 120 | |
| done | |
| echo "=== $name (embodiment $EMB) $(date +%H:%M:%S)" | |
| rm -rf data/shards_tmp/$name | |
| python $S/build_shards.py --repo-id unitreerobotics/$name --skip-bad-episodes \ | |
| --out data/shards_tmp/$name --hz 10 --cache-dir .hf_home/tmpcache_$name \ | |
| && python $S/pack_wds.py --src data/shards_tmp/$name \ | |
| --out data/wds/unitree/$name --chunk 50 --maxcount 3000 --embodiment-id $EMB \ | |
| && python $S/validate_wds.py data/wds/unitree/$name --batches 0 \ | |
| || { echo "FAILED $name"; rm -rf data/wds/unitree/$name; } | |
| rm -rf data/shards_tmp/$name .hf_home/tmpcache_$name | |
| } | |
| export -f one | |
| export S HF_HOME | |
| printf "%s\n" $UNITREE_G1 $UNITREE_Z1 | xargs -P 4 -I{} bash -c 'one {}' 2>&1 | |
| python $S/hf_push_wds.py --src data/wds/unitree --repo AlexWortega/wds-unitree | |
| echo "=== UNITREE DONE $(date +%H:%M:%S)" | |