# jax-ml/scaling-book

Home for "How To Scale Your Model", a short blog-style textbook about scaling LLMs on TPUs

Repository: https://github.com/jax-ml/scaling-book
Canonical: https://ross.abutalabs.com/products/scaling-book
Homepage: https://jax-ml.github.io/scaling-book/
Language: HTML
License: MIT
License Family: permissive
Topics: jax, llm-inference, llms, roofline, tpus
Last push: 2026-08-20T16:07:33+00:00

## Health v2 (maintenance only)
Score: 65/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 35, longevity 41
- inputs: {"age_days": 577, "days_push": 13, "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 1371, forks 194 (observed 2026-08-28T04:04:31.934609+00:00)

## What it is
An open-source, blog-style online textbook called 'How To Scale Your Model' that explains how TPUs and GPUs work and how to scale LLM training and inference efficiently. It is built as a Jekyll site with interactive Jupyter-based content, published via GitHub Pages.

## Use cases
- learn how TPUs work and how LLMs run at scale
- understand roofline analysis for accelerator performance
- choose parallelism schemes for LLM training and inference
- estimate the cost and time to train or serve a large Transformer model
- understand collective communication operations like AllGather
- learn how to avoid communication bottlenecks when scaling models

## When to choose
- you want a free, in-depth systems-level explanation of LLM scaling on TPUs/GPUs
- you need to reason about parallelism strategies (data, tensor, pipeline, etc.) for large models
- you want to ballpark training/inference costs and memory requirements
- you have basic LLM and JAX familiarity and want to go deeper on hardware efficiency

## When to avoid
- you need hands-on training code or a runnable framework rather than a textbook
- you are a complete beginner with no understanding of Transformers or LLM training
- you need vendor-specific GPU (CUDA) optimization guides rather than TPU-focused systems theory

## Facets
- artifact type: learning-resource
- maturity: active
- function: documentation, developer-tools
- domain: large-language-models, gpu-computing, tutorials, performance, machine-learning
- platform: python, ruby
- tags: tpu, llm-training, llm-inference, roofline-analysis, parallelism, jax, textbook, distill-style, web-server

## Member repositories
- jax-ml/scaling-book (main) score 65

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:31.934609+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 2026-08-30T04:40:50.550522+00:00, confidence not recorded.
  - readme: https://github.com/jax-ml/scaling-book (fetched 2026-08-28T04:04:31.934609+00:00, sha 952c3db0f01c)
  - homepage: https://jax-ml.github.io/scaling-book/ (fetched 2026-08-29T11:57:28.457202+00:00, sha e82f03144861)
- Data as of 2026-08-30T08:39:29.467469+00:00.