Skip to content
Robotics

twanghcmut/molmoact2-so101-left-arm-task3

robot arm control model

2 min readupdated

This model is a fully fine-tuned version of the MolmoAct2 architecture designed to control the left arm of an SO101 robot. The checkpoint combines three different object manipulation tasks, and the desired behavior is selected using a language instruction. Created using the LeRobot library, the model targets practical applications in robotics and autonomous manipulators, making it useful for researchers and developers working with robotic hardware and VLA (Vision-Language-Action) algorithms.

Key features

  • Trained on three combined cube and box manipulation tasks.
  • Uses the molmoact2 policy from the LeRobot ecosystem.
  • Outputs absolute joint targets in chunks of 30 steps.
  • Accepts image inputs from two cameras (main overhead and wrist cameras).

What you can do

  • Run control of the SO101 left arm in simulation or on real hardware.
  • Test executing cube-moving and drawer-opening tasks using text prompts.
  • Integrate the model into control pipelines via the LeRobot or VLA-RT libraries.

Example

from lerobot.policies.molmoact2.modeling_molmoact2 import MolmoAct2Policy
from lerobot.policies.factory import make_pre_post_processors
from lerobot.configs.policies import PreTrainedConfig

repo = "twanghcmut/molmoact2-so101-left-arm-task3"
policy = MolmoAct2Policy.from_pretrained(repo).to("cuda").eval()
pre, post = make_pre_post_processors(PreTrainedConfig.from_pretrained(repo), pretrained_path=repo)

batch = {
    "observation.images.top":   top,     # (1, 3, 360, 640) float in [0, 1]
    "observation.images.wrist": wrist,   # (1, 3, 480, 640) float in [0, 1]
    "observation.state":        state,   # (1, 6) float32, raw joint units
    "task": "Take the red cube out of the steel pot and place it into the orange hollow cube.",
}
actions = post(policy.predict_action_chunk(pre(batch)))  # (1, 30, 6) absolute joint targets
Read moreDetails
• Base model: allenai/MolmoAct2-SO100_101 (full fine-tune, frozen embeddings)
• Training: 24,000 steps, batch size 32, 1x H100 80GB (took 23.7 hours)
• Limitations: 159 episodes used entirely for training (no held-out split), specific drawer-opening task has a higher error rate
• Data: 30 fps, state and action dimensionality 6 (shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll + gripper)
Discuss Telegram 0 8

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 11 GB · ready files13 GB
  • FP8 / INT8recommendedweights 5.1 GB6.1 GB
  • 4-bitweights 2.9 GB3.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 5060 Ti 8GB · BF16 — from RTX 5080

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # twanghcmut/molmoact2-so101-left-arm-task3
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download twanghcmut/molmoact2-so101-left-arm-task3 --local-dir ~/models/molmoact2-so101-left-arm-task3
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of MolmoAct2-SO100_101 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 1All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

twanghcmut/molmoact2-so101-left-arm-task3
Author
twanghcmut
Category
Robotics
Released
Posted

On NeyroLearn

Hugging Face

80

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments