# facebookresearch/metaseq

Repo for external large-scale work

Repository: https://github.com/facebookresearch/metaseq
Canonical: https://ross.abutalabs.com/products/metaseq
Language: Python
License: MIT
License Family: permissive
Archived: true
Last push: 2024-04-27T22:15:40+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 1584, "days_push": 858, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, archived
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 6548, forks 714 (observed 2026-08-28T04:09:45.138716+00:00)

## What it is
Metaseq is a PyTorch codebase from Meta AI for training and working with large-scale Open Pre-trained Transformers (OPT), forked from fairseq. It provides training workflows and APIs for models up to 175B parameters, with community integrations for inference via Hugging Face, DeepSpeed, FasterTransformer, and others.

## Use cases
- train large language models on hundreds of GPUs
- reproduce or extend the OPT-175B training pipeline
- fine-tune large transformer models with data and model parallelism
- convert OPT checkpoints for use with inference engines like FasterTransformer or CTranslate2
- study the training log of a 175B parameter model
- run large-scale pretraining experiments in PyTorch

## When to choose
- you need to pretrain or fine-tune very large transformer models with distributed training
- you want to work directly with the OPT model family in its original codebase
- you need a fairseq-compatible codebase for large-scale LLM research

## When to avoid
- you just want to run inference with OPT models - use Hugging Face Transformers instead
- you need actively maintained tooling - the repo is largely in maintenance mode
- you are training small models where simpler frameworks like Hugging Face Trainer suffice

## Facets
- artifact type: library
- maturity: maintenance
- function: llm-training, llm-inference, deep-learning, machine-learning
- domain: large-language-models, deep-learning, machine-learning, artificial-intelligence
- platform: python
- tags: opt, transformers, large-scale-training, fairseq-fork, facebook-research, distributed-training, gpu, linux, docker

## Member repositories
- facebookresearch/metaseq (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:45.138716+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 2026-08-29T17:44:04.806151+00:00, confidence not recorded.
  - readme: https://github.com/facebookresearch/metaseq (fetched 2026-08-28T04:09:45.138716+00:00, sha b4329050a0ed)
- Data as of 2026-08-30T08:39:29.467469+00:00.