phospho-app / act-policy

Let's you train an ACT policy from a LeRobot dataset

  • Public
  • 1 run
  • T4

Input

*string
Shift + Return to add a new line

Hugging Face dataset ID to train on, LeRobot format > v2.0 expected, i.e. 'LegrandFrederic/dual-setup'

*string
Shift + Return to add a new line

Hugging Face API token (used to upload your trained model to your HF profile), find yours here: https://huggingface.co/settings/token

string
Shift + Return to add a new line

Hugging Face model name, the name of the model to be created, i.e. 'trained-act-replicate'

Default: "trained-act-replicate"

string
Shift + Return to add a new line

Weights & Biases API key (optional, to track the online training), find yours here: https://wandb.ai/authorize

Default: ""

string
Shift + Return to add a new line

Weights & Biases project name

Default: "ACT-Replicate"

string
Shift + Return to add a new line

Name for this training job

Default: "replicate_training"

integer
(minimum: 1, maximum: 200000)

Number of steps to train for

Default: 100000

Output

No output yet! Press "Submit" to start a prediction.

Run time and cost

This model runs on Nvidia T4 GPU hardware. We don't yet have enough runs of this model to provide performance information.

Readme

ACT policy - phospho easy training pipeline

This setup allows you to train an ACT policy, using LeRobot for training.

🚀 What This Pipeline Does

  • Trains ACT models using datasets formatted for LeRobot (v2.0+).
  • Automatically validates dataset availability and access permissions.
  • Uploads trained models to your Hugging Face profile.
  • Supports Weights & Biases for tracking training progress.
  • Runs training efficiently on GPUs with configurable steps and job names.

🔗 Useful Links

🛠 Setup & Usage

To train your model, provide the following inputs:

  • Dataset Repo ID (Hugging Face dataset formatted for LeRobot).
  • Hugging Face Token (for uploading trained models).
  • Model Name (for storing results in Hugging Face).
  • Weights & Biases API Key (optional, for experiment tracking).
  • Training Steps The default is 100K steps, you can leave it like so or lower to 60/80K for quicker and cheaper training.

The pipeline ensures all necessary dependencies are installed, including LeRobot and related libraries, before executing the training process. Once trained, the model will be available on Hugging Face for further use, you will also be able to download a .zip archive of your model.