Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Online Goal-Conditioned Reinforcement Learning in JAX. ICLR 2025 Spotlight.
| Date | Stars |
|---|---|
| 2026-07-31 | 275 |
| 2026-08-02 | 275 |
| 2026-08-03 | 275 |
| 2026-08-05 | 276 |
| 2026-08-06 | 276 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<h1 align="center"> JaxGCRL</h1>
<p align="center">
<a href= "https://arxiv.org/abs/2408.11052">
<img src="https://img.shields.io/badge/arXiv-2311.10090-b31b1b.svg" /></a>
<a href="https://pypi.org/project/jaxgcrl/">
<img src="https://img.shields.io/pypi/v/jaxgcrl" /></a>
<a href= "https://github.com/MichalBortkiewicz/JaxGCRL/blob/master/LICENSE">
<img src="https://img.shields.io/badge/license-Apache2.0-blue.svg" /></a>
<a href= "https://michalbortkiewicz.github.io/JaxGCRL/">
<img src="https://img.shields.io/badge/docs-green" /></a>
<a href= "https://michalbortkiewicz.github.io/JaxGCRL/">
<img src="https://img.shields.io/badge/website-purple" /></a>
</p>
<p align="center"><img src="https://raw.githubusercontent.com/MichalBortkiewicz/JaxGCRL/master/imgs/grid_transparent.png" width=85%></p>
<center>
[**Installation**](#Installation) | [**Quick Start**](#start) | [**Environments**](#envs) | [**Baselines**](#baselines) | [**Citation**](#cite)
</center>
<br/>
## Accelerating Goal-Conditioned RL Algorithms and Research
arXiv link: [https://arxiv.org/abs/2408.11052](https://arxiv.org/abs/2408.11052)
We provide blazingly fast goal-conditioned environments based on [MJX](https://mujoco.readthedocs.io/en/stable/mjx.html) and [BRAX](https://github.com/google/brax) for
quick experimentation with goal-conditioned self-supervised reinforcement learning.
- **Blazingly Fast Training** - Train 10 million environment steps in 10
minutes on a single GPU, up to $22\times$ faster than prior implementations.
- **Comprehensive Benchmarking** - Includes 10+ diverse environments and multiple pre-implemented baselines for out-of-the-box evaluation.
- **Modular Implementation** - Designed for clarity and scalability,
allowing for easy modification of algorithms.
## Installation 📂
#### Editable Install (Recommended)
After cloning the repository, run one of the following commands.
With GPU on Linux:
```bash
pip install -e . -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
```
> [!NOTE]
> Make sure you have the correct CUDA version installed, i.e. CUDA >= 12.3.
> You can check your CUDA version with `nvcc --version` command.
> If you have an older version, you can create a new conda environment with the correct version of CUDA and JaxGCRL package using the following command:
> ```bash
> conda env create -f environment.yml
> ```
With CPU on Mac:
```bash
export SDKROOT="$(xcrun --show-sdk-path)" # may be needed to build brax dependencies
pip install -e .
```
#### PyPI
The package is also available on PyPI:
```bash
pip install jaxgcrl -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
```
<h3 name="start" id="start">Quick Start 🚀 </h3>
To verify the installation, run a test experiment:
```bash
jaxgcrl crl --env ant
```
The `jaxgcrl` command is equivalent to invoking `python run.py` with the same arguments
> [!NOTE]
> If you haven't yet configured [`wandb`](https://wandb.ai/site), you may be prompted to log in.
See `scripts/train.sh` for an example config.
A description of the available agents can be generated with `jaxgcrl --help`.
Available configs can be listed with `jaxgcrl {crl,ppo,sac,td3} --help`.
Common flags you may want to change include:
- **env=...**: replace "ant" with any environment name. See `jaxgcrl/utils/env.py` for a list of available environments.
- Removing **--log_wandb**: omits logging, if you don't want to use a wandb account.
- **--total_env_steps**: shorter or longer runs.
- **--num_envs**: based on how many environments your GPU memory allows.
- **--contrastive_loss_fn, --energy_fn, --h_dim, --n_hidden, etc.**: algorithmic and architectural changes.
> [!Note]
> We recommend using [calculator by @riiswa](https://warisradji.com/jaxgcrl-calculator/) for checking the correctness of hyperparameters:
> <p align="center">
> <img src="https://raw.githubusercontent.com/MichalBortkiewicz/JaxGCRL/master/Excerpt of 14,482 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:91ccb9f3597781bf, llm:Repository topics: goal-conditioned-rl, jax, reinforcement-learning, unsupervised-reinforcement-learning; description: Online Goal-Conditioned Reinforcement Learning in JAX (ICLR 2025 Spotlight).
matched fp:91ccb9f3597781bf, llm:Repository topics: goal-conditioned-rl, jax, reinforcement-learning, unsupervised-reinforcement-learning; description: Online Goal-Conditioned Reinforcement Learning in JAX (ICLR 2025 Spotlight).
matched fp:91ccb9f3597781bf, llm:Repository topics: goal-conditioned-rl, jax, reinforcement-learning, unsupervised-reinforcement-learning; description: Online Goal-Conditioned Reinforcement Learning in JAX (ICLR 2025 Spotlight).