# PiotrNawrot/nanoT5

Fast & Simple repository for pre-training and fine-tuning T5-style models

Repository: https://github.com/PiotrNawrot/nanoT5
Canonical: https://ross.abutalabs.com/products/nanot5
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2024-08-21T08:13:55+00:00

## Health v2 (maintenance only)
Score: 21/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 8, longevity 91
- inputs: {"age_days": 1274, "days_push": 742, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1021, forks 78 (observed 2026-08-28T04:03:15.926590+00:00)

## What it is
nanoT5 is a PyTorch codebase for pre-training and fine-tuning T5-style encoder-decoder language models on a limited budget, e.g. a single A100 GPU in under 24 hours. It serves as an accessible, reproducible research template and educational baseline, complementing nanoGPT and Cramming.

## Use cases
- pre-train a T5 model from scratch on a single GPU
- reproduce T5 v1.1 pre-training in PyTorch
- fine-tune an encoder-decoder model on Super-Natural Instructions
- learn how LLM pre-training pipelines work
- get a starting template for NLP research experiments
- compare against nanoGPT for encoder-decoder architectures

## When to choose
- you want to pre-train or fine-tune a T5-style model with limited compute
- you need a reproducible, simple PyTorch baseline for encoder-decoder LLM research
- you want to learn the end-to-end pre-training pipeline

## When to avoid
- you need production-scale distributed training on TPU pods
- you want a decoder-only (GPT-style) model - use nanoGPT instead
- you just need to use an off-the-shelf pretrained T5 from HuggingFace

## Facets
- artifact type: learning-resource
- maturity: maintenance
- function: llm-training, machine-learning, nlp
- domain: large-language-models, deep-learning, tutorials
- platform: python
- tags: t5, encoder-decoder, pre-training, pytorch, research-template, nano-models, fine-tuning, natural-language-processing, gpu, linux

## Member repositories
- PiotrNawrot/nanoT5 (main) score 21

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:15.926590+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 2026-08-30T07:08:48.584063+00:00, confidence not recorded.
  - readme: https://github.com/PiotrNawrot/nanoT5 (fetched 2026-08-28T04:03:15.926590+00:00, sha 09cc90f44576)
- Data as of 2026-08-30T08:39:29.467469+00:00.