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

# g1_fdmv2_broccoli_500step — LingBot-VA G1 post-trained transformer

Fine-tuned `transformer` for LingBot-VA on Unitree G1 (Dex1), task
`yigao7117/put_cup_n_broccoli`:
*"Pick the pink object and put it in the orange basket, then pick up the
broccoli and put it inside the pink object."*

- Base: `robbyant/lingbot-va-base`
- Post-training: 50 demos, single task, lr 1e-5, **FDM v2 recipe** — the
  mutually-exclusive per-microstep regime (rank-synced coin `fdm_prob=0.5`:
  EITHER FDM video-only L_fdm Eq.13 `lambda_fdm=1.0` OR standard IDM
  L_dyn+L_inv; one forward, one backward). Optimizer **step 500** of a 2000-step
  run.
- 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 armanakbari4/g1_fdmv2_broccoli_500step --local-dir g1_broc_500_dl

mkdir -p g1_broc_500
ln -sf $(realpath g1_broc_500_dl/transformer)  g1_broc_500/transformer
ln -sf $(realpath lingbot-va-base/vae)          g1_broc_500/vae
ln -sf $(realpath lingbot-va-base/text_encoder) g1_broc_500/text_encoder
ln -sf $(realpath lingbot-va-base/tokenizer)    g1_broc_500/tokenizer
```

Serve with `CONFIG_NAME=g1_cupbroc MODEL_PATH=g1_broc_500`.
`transformer/config.json` has `attn_mode: torch` (inference-ready).