armanakbari4 commited on
Commit
c827be7
·
verified ·
1 Parent(s): 6119088

g1 cup_broccoli FDM-v2 transformer @ step 500

Browse files
README.md ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ tags:
4
+ - robotics
5
+ - lingbot-va
6
+ - unitree-g1
7
+ - world-model
8
+ ---
9
+
10
+ # g1_fdmv2_broccoli_500step — LingBot-VA G1 post-trained transformer
11
+
12
+ Fine-tuned `transformer` for LingBot-VA on Unitree G1 (Dex1), task
13
+ `yigao7117/put_cup_n_broccoli`:
14
+ *"Pick the pink object and put it in the orange basket, then pick up the
15
+ broccoli and put it inside the pink object."*
16
+
17
+ - Base: `robbyant/lingbot-va-base`
18
+ - Post-training: 50 demos, single task, lr 1e-5, **FDM v2 recipe** — the
19
+ mutually-exclusive per-microstep regime (rank-synced coin `fdm_prob=0.5`:
20
+ EITHER FDM video-only L_fdm Eq.13 `lambda_fdm=1.0` OR standard IDM
21
+ L_dyn+L_inv; one forward, one backward). Optimizer **step 500** of a 2000-step
22
+ run.
23
+ - This repo contains **only `transformer/`** — `vae/`, `text_encoder/`,
24
+ `tokenizer/` are unchanged from `robbyant/lingbot-va-base`.
25
+
26
+ ## Assemble an eval-ready checkpoint
27
+
28
+ ```bash
29
+ hf download robbyant/lingbot-va-base --local-dir lingbot-va-base
30
+ hf download armanakbari4/g1_fdmv2_broccoli_500step --local-dir g1_broc_500_dl
31
+
32
+ mkdir -p g1_broc_500
33
+ ln -sf $(realpath g1_broc_500_dl/transformer) g1_broc_500/transformer
34
+ ln -sf $(realpath lingbot-va-base/vae) g1_broc_500/vae
35
+ ln -sf $(realpath lingbot-va-base/text_encoder) g1_broc_500/text_encoder
36
+ ln -sf $(realpath lingbot-va-base/tokenizer) g1_broc_500/tokenizer
37
+ ```
38
+
39
+ Serve with `CONFIG_NAME=g1_cupbroc MODEL_PATH=g1_broc_500`.
40
+ `transformer/config.json` has `attn_mode: torch` (inference-ready).
transformer/config.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "patch_size": [
3
+ 1,
4
+ 2,
5
+ 2
6
+ ],
7
+ "num_attention_heads": 24,
8
+ "attention_head_dim": 128,
9
+ "in_channels": 48,
10
+ "out_channels": 48,
11
+ "action_dim": 30,
12
+ "text_dim": 4096,
13
+ "freq_dim": 256,
14
+ "ffn_dim": 14336,
15
+ "num_layers": 30,
16
+ "cross_attn_norm": true,
17
+ "eps": 1e-06,
18
+ "rope_max_seq_len": 1024,
19
+ "pos_embed_seq_len": null,
20
+ "attn_mode": "torch",
21
+ "_class_name": "WanTransformer3DModel",
22
+ "_diffusers_version": "0.35.0.dev0"
23
+ }
transformer/diffusion_pytorch_model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:55b9131338da8b831334f35f9289c91b52c089ef2c2996e1a59efd136032eeb6
3
+ size 10177831668