README.md

---
license: apache-2.0
tags:
- robotics
- lingbot-va
- unitree-g1
- world-model
---

# UnitreeG1_putawaytoolsV2_rndchnk_2000step — LingBot-VA G1 post-trained transformer

Fine-tuned `transformer` for LingBot-VA on Unitree G1 (Dex1), task
`XiaoweiLinXL/pi05-unitree-g1-put-away-tools-v2.1` (cleaner recollection
of the original `put_away_tools` task):
*"Put the battery on the shelf labeled 'battery' and put the screwdriver
on the shelf labeled 'Philips'."*

- Base: `robbyant/lingbot-va-base`
- Post-training: **70 demos (43,851 frames)** — cleaner than the v1 release
  (48 demos / 24,271 frames). Single-task, lr 1e-5, **FDM v2 recipe** —
  mutually-exclusive per-microstep regime (rank-synced coin `fdm_prob=0.5`:
  FDM video-only L_fdm Eq.13 `lambda_fdm=1.0` OR standard IDM L_dyn+L_inv;
  one forward, one backward). Per-step **randomized chunk_size ∈ {1,2,3,4}**
  and **window_size ∈ {4..64}** so the deployed model handles any chunk/window
  setting at inference (the "rndchnk" in the repo name).
- 4 GPUs × `grad_accum=4` = effective batch 16, optimizer **step 2000** of a
  5000-step schedule.
- Action range under quantile normalization: **±1.26** (vs the v1
  data's ±2.0 — the cleaner demos hug the q01/q99 quantile range much
  more tightly, removing the out-of-range tail that hurt v1 deployment).
- This repo contains **only `transformer/`** — `vae/`, `text_encoder/`,
  `tokenizer/` are unchanged from `robbyant/lingbot-va-base`.

## Assemble an eval-ready checkpoint

```bash
hf download robbyant/lingbot-va-base                              --local-dir lingbot-va-base
hf download EmbodyX/UnitreeG1_putawaytoolsV2_rndchnk_2000step      --local-dir g1_pat_v2_2000_dl

mkdir -p g1_pat_v2_2000
ln -sf $(realpath g1_pat_v2_2000_dl/transformer)  g1_pat_v2_2000/transformer
ln -sf $(realpath lingbot-va-base/vae)            g1_pat_v2_2000/vae
ln -sf $(realpath lingbot-va-base/text_encoder)   g1_pat_v2_2000/text_encoder
ln -sf $(realpath lingbot-va-base/tokenizer)      g1_pat_v2_2000/tokenizer
```

Serve with `CONFIG_NAME=g1_putawaytools_v21 MODEL_PATH=g1_pat_v2_2000`.
`transformer/config.json` has `attn_mode: torch` (inference-ready).