# tinkoff-ai/CORL

High-quality single-file implementations of SOTA Offline and Offline-to-Online RL algorithms: AWAC, BC, CQL, DT, EDAC, IQL, SAC-N, TD3+BC, LB-SAC, SPOT, Cal-QL, ReBRAC

Repository: https://github.com/tinkoff-ai/CORL
Canonical: https://ross.abutalabs.com/products/corl
Homepage: https://arxiv.org/abs/2210.07105
Language: Python
License: Apache-2.0
License Family: permissive
Topics: d4rl, gym, offline-reinforcement-learning, reinforcement-learning
Archived: true
Last push: 2023-08-03T07:13:55+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 1440, "days_push": 1126, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: archived
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1372, forks 177 (observed 2026-08-28T04:04:32.349175+00:00)

## What it is
CORL is a research-oriented Python library providing thoroughly benchmarked single-file implementations of state-of-the-art offline and offline-to-online reinforcement learning algorithms such as CQL, IQL, AWAC, DT, and ReBRAC. It emphasizes readable, research-friendly code with Weights & Biases experiment tracking and benchmarked results on D4RL datasets.

## Use cases
- implement offline RL algorithms like CQL or IQL from a single readable file
- benchmark offline RL methods on D4RL datasets
- run offline-to-online fine-tuning experiments
- reproduce published offline RL results with transparent hyperparameters
- track and log RL experiments with Weights and Biases
- study performance-relevant implementation details of ORL algorithms

## When to choose
- you need clean, single-file reference implementations of offline RL algorithms
- you want benchmarked, reproducible results on D4RL
- you are doing offline RL research and need to tune or extend algorithms
- you want experiment tracking integrated out of the box

## When to avoid
- you need online RL algorithms (use CleanRL instead)
- you need discrete-control offline RL (see the authors' Katakomba library)
- you want a production RL training framework with modular abstractions
- you need actively maintained code with frequent updates

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, reinforcement-learning, benchmarking
- domain: reinforcement-learning, machine-learning, artificial-intelligence
- platform: python
- tags: offline-rl, offline-to-online-rl, d4rl, single-file-implementations, wandb, cleanrl-inspired, research-codebase, gpu, docker

## Member repositories
- tinkoff-ai/CORL (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:32.349175+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:40:48.054398+00:00, confidence not recorded.
  - readme: https://github.com/tinkoff-ai/CORL (fetched 2026-08-28T04:04:32.349175+00:00, sha 51d5a09512b3)
  - homepage: https://arxiv.org/abs/2210.07105 (fetched 2026-08-29T11:57:30.822194+00:00, sha 4cbd1253c953)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T11:57:30.824875+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T11:57:30.828358+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T11:57:30.830567+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T11:57:30.826619+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.