Downloads · 30 days
40
36% of all-time downloads
molmoact2_Reasoning_Step_076596
justintiensmith/molmoact2_Reasoning_Step_076596
molmoact2_Reasoning_Step_076596 is a robotics model from justintiensmith. Use it for the robotics task on the model card, and read the license before you ship it in a product. It is set up for lerobot.
This is the MolmoAct2 checkpoint evaluated in Spa-Bench, a real-robot benchmark of spatially grounded reasoning in vision-language-action policies.
All-time downloads
112
Public
Parameters
5.4B
10.9 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors10.9 GB · 100%
Parameter types
How the weights are stored.
BF165.4B · 100%
Base models
Model card
From the Hugging Face model README
MolmoAct2 — Spa-Bench Epoch 12 (Step 76,596)
This is the MolmoAct2 checkpoint evaluated in Spa-Bench, a real-robot benchmark of spatially grounded reasoning in vision-language-action policies.
Model details
| Field | Value |
|---|---|
| Model repository | justintiensmith/molmoact2_Reasoning_Step_076596 |
| Base model | allenai/MolmoAct2 at e432d85f6e039edca44afb93c262f3084ab72a9c |
| Checkpoint | End of epoch 12; step 76,596 |
| Robot | SO-101 single-arm manipulator |
| Inputs | Fixed middle RGB, wrist RGB, six absolute joint positions, text instruction |
| Outputs | Six absolute joint-position targets |
| Action horizon | 30 |
| Normalization | Min–max normalization for state and action |
| Adaptation | VLM and continuous-action expert updated; token embeddings frozen |
| Optimizer | AdamW; zero weight decay |
| Learning rates | 1e-5 base, 5e-6 vision/connector, 5e-5 action expert |
| Schedule | 200 warm-up updates, then cosine decay |
| Hardware and batch | Four NVIDIA GH200 GPUs; 24 samples per device, global batch 96 |
The embedded train_config.json pins the training data to
justintiensmith/VLA_Reasoning_Training_Dataset_1200@3c92bdd.
That dataset contains 1,200 episodes, 612,733 frames, 321 instruction strings,
and five RGB views; this policy consumes only the middle and wrist views.
The author-supplied original
train_molmoact2_v7_isambard.sh
launcher is archived with the thesis artifact. It records the Isambard
environment, pinned base revision, seed 1000, four-GPU launch, dataset pin,
checkpoint cadence, and resume behavior used for this run.
Physical evaluation
| Condition | Successes | Rate |
|---|---|---|
| Familiar/in-distribution spatial instructions | 87/120 | 72.5% |
| All withheld/OOD spatial configurations | 156/300 | 52.0% |
| Matched OOD subset | 73/120 | 60.8% |
| Matched direct-manipulation controls | 107/120 | 89.2% |
The matched-control gap was 28.3 percentage points. These are physical rollout results, not simulation metrics.
- Rollouts:
justintiensmith/Spa_Bench_Full_MolmoAct2 - Thesis artifact:
justintiensmith/Imperial-Thesis
Intended use and limitations
This release supports reproduction and analysis of the Spa-Bench experiment. The results apply to this checkpoint, SO-101 embodiment, workspace, cameras, objects, and protocol; they do not establish general robot capability. No held-out validation loss was used to select among checkpoints: epoch 12 was fixed before physical evaluation.
Robot policies can move hardware unexpectedly. Use conservative motion limits, an accessible emergency stop, a clear workspace, and direct supervision. Do not deploy this checkpoint for unattended or safety-critical operation.
The repository does not relicense third-party components. Use is subject to the terms of the MolmoAct2 base model, LeRobot software, and released data.
Citation
Please cite the completed Spa-Bench MSc report, the thesis artifact, and the MolmoAct2 work referenced in the report.
- Task
- Robotics
- Library
- lerobot
- Languages
- en
- Created
- Aug 10, 2026
- Updated
- Sep 9, 2026