AI Potluck
Model components / Fine-tuning code

Atropos

Nous Research

Language-model reinforcement-learning environments framework for collecting and evaluating LLM trajectories across diverse environments. Nous positions it as part of the post-training stack, and the GitHub repo has 1.2K stars and 362 forks. The repo was archived (read-only) on 2026-07-04 and is no longer actively developed.

Atropos, Nous Research LLM RL-environments framework for collecting/evaluating trajectories and driving RL training (RLHF/RLAIF, GRPO-style). v0.4.0 (Mar 10, 2026), confirmed live on GitHub June 2026. Integrates with trainers Axolotl and Tinker. Borderline: it is the RL-environment/orchestration layer that feeds a fine-tuning trainer. Repo archived read-only on 2026-07-04 (last release v0.4.0, Mar 10 2026); no longer actively maintained.

Openness

5 high confidence
5.0
license
MIT(OSI)
source
public
scope
async RL-environments framework + trajectory API
trainer integrations
Axolotl/Tinker
core-gated
ungated

Fully OSI-licensed (MIT) and public, no managed/closed core. Fits finetuning_code as the RL-training-environment layer (RLHF/RLAIF/GRPO).

  • https://github.com/NousResearch/atropos recorded 2026-07-21

    MIT license; LLM RL-environments framework; v0.4.0 (Mar 10 2026); Axolotl/Tinker trainer integrations; repo archived read-only 2026-07-04

Adoption

1 low confidence
1.0

No PyPI/download or named-user count found on the primary channel; only 1.3k GitHub stars as a signal (stars_fallback => capped <=3, placed at 1). Early/niche research adoption.

Capability

3 medium confidence
3.0

Capable, well-architected RL-environments layer; mid-tier as a fine-tuning enabler (it orchestrates RL rather than implementing large-scale training itself). Vendor-reported gains (4.6x parallel-task, 2.5x finance) not independently benchmarked.

Unchanged since 2026-07-30 (last edited, not re-checked)