# Xwin-LM/Xwin-LM

Xwin-LM: Powerful, Stable, and Reproducible LLM Alignment

Repository: https://github.com/Xwin-LM/Xwin-LM
Canonical: https://ross.abutalabs.com/products/xwin-lm
Language: Python
License Family: other
Last push: 2024-05-31T11:26:15+00:00

## Health v2 (maintenance only)
Score: 28/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 35, longevity 77
- inputs: {"age_days": 1087, "days_push": 824, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1035, forks 45 (observed 2026-08-28T04:03:18.918850+00:00)

## What it is
Xwin-LM is an open-source project for LLM alignment technologies including supervised fine-tuning, reward models, reject sampling, and RLHF (PPO), built on Llama base models. It also releases fine-tuned models for general chat, math reasoning, and code generation that topped benchmarks like AlpacaEval and MATH.

## Use cases
- train an LLM with RLHF and PPO
- fine-tune Llama with supervised fine-tuning
- train a reward model for alignment
- improve model math reasoning on GSM8K and MATH
- reproduce top AlpacaEval open-source chat model
- align a code generation model on HumanEval

## When to choose
- you want a full alignment pipeline (SFT, RM, PPO) for Llama-based models
- you need reproducible RLHF training code with strong benchmark results
- you want proven fine-tuned checkpoints for chat, math, or code

## When to avoid
- you need inference/serving rather than training
- you need a permissively licensed codebase - the repo has no license
- you need actively maintained tooling - releases have slowed since 2024

## Facets
- artifact type: library
- maturity: maintenance
- function: llm-training, machine-learning, deep-learning
- domain: large-language-models, machine-learning, artificial-intelligence
- platform: python
- tags: rlhf, alignment, sft, reward-model, ppo, llama, reinforcement-learning, huggingface, gpu, linux

## Member repositories
- Xwin-LM/Xwin-LM (main) score 28

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:18.918850+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 2026-08-30T07:04:45.089655+00:00, confidence not recorded.
  - readme: https://github.com/Xwin-LM/Xwin-LM (fetched 2026-08-28T04:03:18.918850+00:00, sha 627417c2698b)
- Data as of 2026-08-30T08:39:29.467469+00:00.