#4949th of 68 in Training and fine-tuning
reagent
Archived Meta platform for offline reinforcement learning on PyTorch
- Stars
- 3.7k
- License
- BSD-3-Clause
- Last commit
- Nov 2026
- Language
- Python
3,686 on 25 Apr 20263,727 on 14 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
ReAgent is a Python platform from Meta for applied reinforcement learning, using PyTorch for training and TorchScript for serving. It covers data preprocessing, feature transformation, distributed training and counterfactual policy evaluation, with off-policy algorithms such as DQN variants, SAC, TD3, PPO and bandits. The repository is archived and no longer maintained; the README points to Pearl as the successor.
Who it is for: RL researchers studying offline RL and counterfactual evaluation
Strengths
- Broad algorithm set: DQN variants, TD3, SAC, CRR, PPO, SlateQ, Seq2Slate, bandits
- Counterfactual policy evaluation (Doubly Robust, MAGIC) to test policies without deploying
- Offline batch training workflow with TorchScript model serving
- Includes Domain Analysis Tool and Behavior Cloning for bootstrapping real-world RL
Weaknesses
- Officially archived and unmaintained; README directs users to Pearl
- No tagged releases
- Aimed at large-scale offline recommendation tasks, not simulator-based RL
- Copyright notice is 2022, so dependencies may be outdated
What it needs
- Needs PyTorch, TorchScript
- Models: PyTorch, TorchScript
Also in Training and fine-tuning
See all 68| Rank | Project | Score |
|---|---|---|
| 1 | peftPython library for LoRA and other parameter-efficient fine-tuning of Transformers models | 89 out of 100 |
| 2 | unslothDesktop and web app to run and fine-tune local models | 88 out of 100 |
| 3 | trlPost-training library with SFT, DPO, GRPO and KTO trainers | 88 out of 100 |
| #4 | deepspeedPyTorch library for distributed training of very large models with ZeRO | 87 out of 100 |
| #5 | megatron-lmDistributed training library for large transformer models on NVIDIA GPUs | 86 out of 100 |