train your robotic AI

train your models in hours instead of days with our GPUs

fully compatible with
Hugging FaceLeRobot
Hero Image

How It Works

Training robotics AI models has never been easier. Follow these simple steps to get started.

1

Add Your Access Token

Add your Hugging Face access token to partabot. This allows us to access your dataset if it's private and upload checkpoints after training.
(You only need to do this once, as the access token can be reused for all your future trainings.)

Production HF Key
Added 2 days ago
Get Token
2

Upload Your Dataset

Upload your own robotics dataset to Hugging Face or choose from existing LeRobot datasets. We support both public and private datasets for training your models.

3

Create a Training

Configure your training by selecting the model type (ACT or diffusion), robot arm (SO-100 / Koch), dataset, and training parameters like steps, batch size, workers, and action horizon.

Model Configuration
Model Type:ACT
Robot Arm:SO-100
Training Parameters
Steps:30,000
Batch Size:48
Action Horizon:50
4

Monitor Your Training

Track your training progress in real-time. View checkpoints (saved every 5,000 steps), logs from the training server, and metrics like Loss and Gradient Norm.

Metrics
Training Metrics Dashboard
Checkpoints
Model Checkpoints List
Logs
Training Logs Console
5

Upload to Hugging Face

Select a checkpoint and trigger an upload to Hugging Face in a repository you provide. We'll create the repo if it doesn't exist yet, making it easy to share and use your trained model.

Checkpoint Upload
Checkpoint 25000 (Final)
Uploaded to username/so100-act-model
Your model is now available on Hugging Face!
Hugging FaceLeRobot

Build on Top of LeRobot

partabot is proud to be part of the LeRobot community, embracing and extending Hugging Face's mission to democratize robotics AI.

State-of-the-art machine learning for real-world robotics with PyTorch integration

Pretrained models and datasets with human collected demonstrations ready to use

Focus on imitation learning and reinforcement learning with proven real-world transfer

Growing community support for affordable and capable robots

"
"

LeRobot Mission

Democratize robotics by removing barriers, enabling everyone to contribute and benefit from a shared ecosystem of datasets and pretrained models.

Supported Models

Train state-of-the-art robotics AI models with our platform.

ACT

Action-Conditioned Transformer for robotic control and manipulation tasks.

Available Now
Diffusion

Diffusion models for generating precise robotic movement trajectories.

Available Now
Pi Zero + LoRA

Flow-based diffusion vision-language-action model (VLA)

Coming Soon
Pi Zero Fast

Autoregressive VLA, based on the FAST action tokenizer

Coming Soon
Coming Soon

Model Inference is Almost Here!

Soon you'll be able to run inference with your trained models directly from our platform. Test your models, validate performance, and deploy them seamlessly - all without leaving partabot.

Real-time Testing

Test your trained models instantly with real-time inference capabilities and immediate feedback.

Performance Analytics

Get detailed performance metrics and analytics to validate your model's effectiveness.

Easy Deployment

Deploy your models with one click and integrate them seamlessly into your robotics applications.

Robot Arm Shop

Get the hardware you need to bring your AI models to life. These robot arms are fully compatible with LeRobot and partabot.

SO-ARM101 Leader
SO-ARM101 Leader
New

The SO-ARM101 robotic arm provides an accessible and customizable robotics experience ideal for enthusiasts, educators, students, and makers.

SO-ARM101
SO-ARM101
New

The SO-ARM101 robotic arm provides an accessible and customizable robotics experience ideal for enthusiasts, educators, students, and makers.

Koch v1.1
Koch v1.1

The Koch v1.1 robotic arm provides an affordable, hands-on robotics experience for students, educators, and hobbyists.