# XYZ-AI-Lab/axrl

AxisRL is an agentic RL post-training framework built on SGLang rollout, Megatron training, and real-world agent workflows.

Repository: https://github.com/XYZ-AI-Lab/axrl
Canonical: https://ross.abutalabs.com/products/axrl
Homepage: https://xyz-lab.ai/
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-03T08:58:45+00:00

## Health v2 (maintenance only)
Score: 55/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 95, release rhythm 35, longevity 2
- inputs: {"age_days": 41, "days_push": 30, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1056, forks 24 (observed 2026-08-28T04:03:24.604044+00:00)

## What it is
AxisRL is an agentic reinforcement learning post-training framework for large language models, built on SGLang for high-throughput rollout and Megatron for large-scale distributed training. It coordinates multi-turn agent environments, tool calls, reward collection, weight synchronization, and reproducible debugging within a single system layer.

## Use cases
- post-train an LLM with PPO or GRPO on multi-turn agent trajectories
- run RL training with hundreds-of-turn tool-calling agent workflows
- scale rollout and training across a GPU cluster with SGLang and Megatron
- capture black-box agent environments through an OpenAI-compatible proxy for RL training
- debug rollout-training mismatches with spike replay and routing analysis

## When to choose
- you need agentic multi-turn RL post-training at large parameter scale
- you want SGLang rollout combined with Megatron distributed training in one framework
- you need configurable policy objectives like PPO, GRPO, GSPO, or TOPR
- you require observability and reproducibility for long agent training runs

## When to avoid
- you only need single-turn supervised fine-tuning without RL
- you lack multi-GPU infrastructure or a serving cluster
- you need a simple plug-and-play RLHF library with minimal setup
- you work outside the Python 3.12+ ecosystem

## Facets
- artifact type: framework
- maturity: active
- function: llm-training, agent-framework, llm-inference, machine-learning, gpu-computing
- domain: large-language-models, machine-learning, reinforcement-learning, gpu-computing
- platform: python
- tags: reinforcement-learning, post-training, sglang, megatron, rlhf, agentic-rl, ppo, grpo, rollout, distributed-training, ai-agents, gpu, linux, docker

## Member repositories
- XYZ-AI-Lab/axrl (main) score 55

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:24.604044+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 2026-08-30T06:58:18.032552+00:00, confidence not recorded.
  - readme: https://github.com/XYZ-AI-Lab/axrl (fetched 2026-08-28T04:03:24.604044+00:00, sha cf994d42d581)
  - homepage: https://xyz-lab.ai/ (fetched 2026-08-29T13:00:08.758241+00:00, sha 6af22f440498)
- Data as of 2026-08-30T08:39:29.467469+00:00.