Publish weights and concise inference instructions

README.md

@@ -0,0 +1,48 @@
+---
+license: other
+library_name: pytorch
+pipeline_tag: robotics
+datasets: [XDOF/ABC-130k]
+tags: [long-wam, yam, abc-130k, robotics]
+---
+
+# Long-WAM YAM
+
+A YAM bimanual robot pretraining model trained on **ABC 130K**, with a
+LongLive 2.0 Robot-S video backbone. It predicts actions after imagining future
+video latents.
+
+## Download
+
+```bash
+hf download Efficient-Large-Model/Long-WAM-YAM --local-dir ./weights/Long-WAM-YAM
+```
+
+## Inference
+
+Load `model.pt` with the matching Long-WAM runtime, `config.yaml` and
+`dataset_stats.json`. Use 4 video denoising steps to sigma 0.9, followed by
+10 action denoising steps.
+
+The three RGB views use a fixed **384 × 320** layout:
+
+```text
++---------------------------------+
+|       Top view: 256 × 320        |
++----------------+----------------+
+| Left wrist     | Right wrist    |
+| 128 × 160      | 128 × 160      |
++----------------+----------------+
+```
+
+P48 covers **1.6 seconds at 30 Hz**. Sample history every four control steps:
+13 images become four clean latent frames, followed by two imagined latents.
+Predict a 32-step action chunk, approximately 1.07 seconds.
+
+State and action are 14-D: left arm's six joints and gripper, then right arm's
+six joints and gripper. Use the supplied z-score statistics to normalize state
+and denormalize predicted actions; RGB inputs use `[-1, 1]`.
+
+The matching runtime and base-model VAE/text assets are required separately.
+This is a pretraining checkpoint: task-specific adaptation and controlled
+closed-loop validation are required before real-robot use.

SHA256SUMS

@@ -0,0 +1,4 @@
+d759778c060040f560d17bee3bbea6df3a2184099b3089f9e7d8b4940dbd55b7  model.pt
+4a84a914bcf939c8240e5ff54cc1dca3006dff912231333406ca099cd449eb83  README.md
+189c1b3c2e229cf4b069b6e85e9f30b894fca72d711b6d8aafb3f71d03b36343  config.yaml
+91f295bf93294214ebc83ace488e08816d29ac93b61d71d70e4ffe22f74fed9b  dataset_stats.json

config.yaml

@@ -0,0 +1,172 @@
+# Inference-only configuration. Load model.pt strictly after model construction.
+# Supply matching text embeddings, or enable the compatible text encoder.
+checkpoint: ./model.pt
+mixed_precision: bf16
+eval_num_inference_steps: 10
+data:
+  train:
+    shape_meta:
+      images:
+      - key: top
+        raw_shape:
+        - 3
+        - 224
+        - 224
+        shape:
+        - 3
+        - 224
+        - 224
+      - key: left_wrist
+        raw_shape:
+        - 3
+        - 224
+        - 224
+        shape:
+        - 3
+        - 224
+        - 224
+      - key: right_wrist
+        raw_shape:
+        - 3
+        - 224
+        - 224
+        shape:
+        - 3
+        - 224
+        - 224
+      action:
+      - key: default
+        raw_shape: 14
+        shape: 14
+      state:
+      - key: default
+        raw_shape: 14
+        shape: 14
+    num_frames: 81
+    past_obs_size: 48
+    action_chunk: 32
+    action_video_freq_ratio: 4
+    video_size:
+    - 384
+    - 320
+    concat_multi_camera: robotwin
+    context_len: 128
+    processor:
+      _target_: fastwam.datasets.lerobot.processors.fastwam_processor.FastWAMProcessor
+      shape_meta:
+        images:
+        - key: top
+          raw_shape:
+          - 3
+          - 224
+          - 224
+          shape:
+          - 3
+          - 224
+          - 224
+        - key: left_wrist
+          raw_shape:
+          - 3
+          - 224
+          - 224
+          shape:
+          - 3
+          - 224
+          - 224
+        - key: right_wrist
+          raw_shape:
+          - 3
+          - 224
+          - 224
+          shape:
+          - 3
+          - 224
+          - 224
+        action:
+        - key: default
+          raw_shape: 14
+          shape: 14
+        state:
+        - key: default
+          raw_shape: 14
+          shape: 14
+      num_obs_steps: 81
+      num_image_steps: 21
+      num_output_cameras: 3
+      action_output_dim: 14
+      proprio_output_dim: 14
+      action_state_transforms: null
+      use_stepwise_action_norm: false
+      norm_default_mode: z-score
+      norm_exception_mode: null
+      action_state_merger:
+        _target_: fastwam.datasets.lerobot.transforms.action_state_merger.ConcatLeftAlign
+      val_transforms:
+      - _target_: fastwam.datasets.lerobot.transforms.image.ToTensor
+    pretrained_norm_stats: ./dataset_stats.json
+model:
+  _target_: fastwam.runtime.create_arwam
+  model_id: Wan-AI/Wan2.2-TI2V-5B
+  tokenizer_model_id: Wan-AI/Wan2.1-T2V-1.3B
+  tokenizer_max_len: 128
+  load_text_encoder: false
+  proprio_dim: 14
+  redirect_common_files: false
+  mot_checkpoint_mixed_attn: true
+  skip_dit_load_from_pretrain: true
+  skip_video_dit_load_from_pretrain: true
+  skip_action_dit_load_from_pretrain: true
+  longlive_video_weights: null
+  action_dit_pretrained_path: null
+  arwam_num_frame_per_block: 1
+  arwam_local_attn_size: -1
+  arwam_sink_size: 0
+  arwam_past_obs_size: 48
+  arwam_num_imagine_frames: 2
+  arwam_imagine_sigma: 0.9
+  arwam_imagine_infer_steps: 4
+  arwam_joint_denoise: false
+  video_dit_config:
+    has_image_input: false
+    patch_size:
+    - 1
+    - 2
+    - 2
+    in_dim: 48
+    hidden_dim: 3072
+    ffn_dim: 14336
+    freq_dim: 256
+    text_dim: 4096
+    out_dim: 48
+    num_heads: 24
+    attn_head_dim: 128
+    num_layers: 30
+    eps: 1.0e-06
+    seperated_timestep: true
+    require_clip_embedding: false
+    require_vae_embedding: false
+    fuse_vae_embedding_in_latents: true
+    use_gradient_checkpointing: true
+    video_attention_mask_mode: per_frame_causal
+    action_conditioned: false
+    action_dim: 14
+    action_group_causal_mask_mode: group_diagonal
+  action_dit_config:
+    action_dim: 14
+    hidden_dim: 1024
+    ffn_dim: 4096
+    num_heads: 24
+    attn_head_dim: 128
+    num_layers: 30
+    text_dim: 4096
+    freq_dim: 256
+    eps: 1.0e-06
+    use_gradient_checkpointing: true
+  video_scheduler:
+    train_shift: 5.0
+    infer_shift: 5.0
+    num_train_timesteps: 1000
+  action_scheduler:
+    train_shift: 5.0
+    infer_shift: 5.0
+    num_train_timesteps: 1000

dataset_stats.json

@@ -0,0 +1,204 @@
+{
+  "action": {
+    "default": {
+      "global_max": [
+        2.7320046424865723,
+        3.2818641662597656,
+        3.204439163208008,
+        2.700853109359741,
+        2.691441297531128,
+        3.7811522483825684,
+        1.0,
+        3.2167577743530273,
+        3.2351346015930176,
+        3.186357259750366,
+        3.09946608543396,
+        2.984330415725708,
+        2.9432990550994873,
+        1.0002068281173706
+      ],
+      "global_mean": [
+        -0.5005107238240027,
+        1.381914052447372,
+        1.0282389744588594,
+        -0.19609857970739863,
+        0.17349895456527187,
+        -0.3958070538075272,
+        0.4321060916782643,
+        0.5512800313762668,
+        1.4804727180244466,
+        1.1117122735819063,
+        -0.2406437357551019,
+        -0.15723303230210917,
+        0.3200179397521419,
+        0.38689656182345383
+      ],
+      "global_min": [
+        -2.662975311279297,
+        -0.378818154335022,
+        -3.0811784267425537,
+        -2.8247804641723633,
+        -2.7046494483947754,
+        -6.9197869300842285,
+        0.0,
+        -2.9336249828338623,
+        -0.35128146409988403,
+        -3.081232786178589,
+        -2.905407667160034,
+        -2.7879931926727295,
+        -3.8900976181030273,
+        0.0
+      ],
+      "global_q01": [
+        -1.0157207326698905,
+        0.7051830743478448,
+        0.407990651693777,
+        -0.9059864495467933,
+        -0.5121502233752125,
+        -1.1316519512331715,
+        0.023071729762461142,
+        -0.07745912643418537,
+        0.7754657879592324,
+        0.4315820502820917,
+        -0.9569596528430835,
+        -0.8317613052450866,
+        -0.4902951630757069,
+        0.01692275123376297
+      ],
+      "global_q99": [
+        0.07271401301061571,
+        2.1668070771412085,
+        1.9000231042792755,
+        0.8037183337660352,
+        0.7994896226295166,
+        0.373592091320015,
+        0.9847720186116629,
+        1.1047793473952356,
+        2.2635719182396654,
+        2.0252360502520084,
+        0.8000953872301054,
+        0.549115486854637,
+        1.0949446105958247,
+        0.9814563312580594
+      ],
+      "global_std": [
+        0.3495856634202851,
+        0.4604627147894431,
+        0.4625276502550232,
+        0.5374316293156617,
+        0.41281121135280635,
+        0.5302864916675254,
+        0.3971947804685133,
+        0.362892788354845,
+        0.43697687766447907,
+        0.4692866453073781,
+        0.5132881050819743,
+        0.4262513416446453,
+        0.5217358174703388,
+        0.3639911014721376
+      ]
+    }
+  },
+  "num_episodes": 129225,
+  "num_transition": 382468339,
+  "state": {
+    "default": {
+      "global_max": [
+        2.722400188446045,
+        3.2164111137390137,
+        3.1466009616851807,
+        1.7015717029571533,
+        1.6321431398391724,
+        2.141031503677368,
+        1.0947552919387817,
+        3.1515603065490723,
+        3.1744487285614014,
+        3.144693613052368,
+        1.8350881338119507,
+        1.6649500131607056,
+        2.1646831035614014,
+        1.0836290121078491
+      ],
+      "global_mean": [
+        -0.5002014805754177,
+        1.3905682227662843,
+        1.0178349787136176,
+        -0.20808514546206375,
+        0.17286741860914037,
+        -0.4002024154821403,
+        0.4714656611298453,
+        0.5501045778928189,
+        1.4886307077258405,
+        1.0994255696016024,
+        -0.2515373244593867,
+        -0.16077766487327597,
+        0.32213089729568706,
+        0.4243088171154238
+      ],
+      "global_min": [
+        -2.593461513519287,
+        -0.004005493130534887,
+        -0.03223468363285065,
+        -1.7092012166976929,
+        -1.6100175380706787,
+        -2.2795071601867676,
+        -0.06222819909453392,
+        -2.141031503677368,
+        -0.005531395319849253,
+        -0.013542382046580315,
+        -1.7130159139633179,
+        -1.659609317779541,
+        -2.1391241550445557,
+        -0.041723888367414474
+      ],
+      "global_q01": [
+        -1.0117017015849312,
+        0.7125901749503004,
+        0.40519094674870604,
+        -0.890344442608663,
+        -0.487476534926118,
+        -1.1169224102203834,
+        0.07559663105985771,
+        -0.07600138597609059,
+        0.7834616794367687,
+        0.42744681026174963,
+        -0.9373523169981716,
+        -0.8116705046539765,
+        -0.47362078209378244,
+        0.06055312201755787
+      ],
+      "global_q99": [
+        0.06982143971411163,
+        2.168851191866642,
+        1.8781019825238348,
+        0.750493507662465,
+        0.7760530405530245,
+        0.3549474786489756,
+        0.9838985491937439,
+        1.100049616703889,
+        2.263974365278712,
+        2.00085369716042,
+        0.7454671833829447,
+        0.5173974011005017,
+        1.079475639591082,
+        0.9805554291104522
+      ],
+      "global_std": [
+        0.3482919988599236,
+        0.46043732106118535,
+        0.4585170040774648,
+        0.5239460726523609,
+        0.4042537307516345,
+        0.5264449965374698,
+        0.3691896556185469,
+        0.3617254804816032,
+        0.4363034754839644,
+        0.46516993227957726,
+        0.4989206819201248,
+        0.4166750697298143,
+        0.5178937337339758,
+        0.3404773192431705
+      ]
+    }
+  }
+}

model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d759778c060040f560d17bee3bbea6df3a2184099b3089f9e7d8b4940dbd55b7
+size 12041813652