# google-research/t5x

Repository: https://github.com/google-research/t5x
Canonical: https://ross.abutalabs.com/products/t5x
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-03T12:03:08+00:00

## Health v2 (maintenance only)
Score: 75/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 95, release rhythm 35, longevity 100
- inputs: {"age_days": 1766, "days_push": 30, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2998, forks 338 (observed 2026-08-28T04:07:37.231057+00:00)

## What it is
T5X is a modular, composable framework built on JAX and Flax for high-performance training, evaluation, and inference of sequence models at many scales. It is a modernized reimplementation of the original T5 codebase, designed for research-friendly, configurable self-service experiments on TPUs and GPUs.

## Use cases
- train sequence-to-sequence models on TPUs
- fine-tune T5-style language models at scale
- run model evaluation and inference for large language models
- run multi-host distributed training experiments on Google Cloud
- reproduce T5 research experiments with a modern JAX stack

## When to choose
- you need to train or fine-tune large sequence models on TPU/GPU hardware
- you want a configurable, research-oriented alternative to the original Mesh TensorFlow T5 codebase
- your workflow already uses JAX and Flax

## When to avoid
- you need a simple PyTorch or TensorFlow training loop for small models
- you want a general-purpose serving/inference server rather than a training framework
- you are not working with sequence models or language tasks

## Facets
- artifact type: framework
- maturity: maintenance
- function: llm-training, machine-learning, deep-learning, benchmarking
- domain: deep-learning, large-language-models, machine-learning
- platform: python, cloud
- tags: jax, flax, tpu, sequence-models, gin-config, google-research, model-training, evaluation, inference, natural-language-processing, gpu, linux

## Member repositories
- google-research/t5x (main) score 75

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:37.231057+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 2026-08-30T07:30:36.919775+00:00, confidence not recorded.
  - readme: https://github.com/google-research/t5x (fetched 2026-08-28T04:07:37.231057+00:00, sha ec1826c18a00)
  - registry_pypi: https://pypi.org/pypi/t5x/json (fetched 2026-08-29T09:45:54.017079+00:00, sha f5b06d334fb7)
- Data as of 2026-08-30T08:39:29.467469+00:00.