# thinking-machines-lab/tinker-cookbook

Post-training with Tinker

Repository: https://github.com/thinking-machines-lab/tinker-cookbook
Canonical: https://ross.abutalabs.com/products/tinker-cookbook
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-26T06:30:36+00:00

## Health v2 (maintenance only)
Score: 85/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 98, longevity 29
- inputs: {"age_days": 415, "days_push": 7, "days_rel": 12, "gap_med": 4.5, "n_releases_24m": 7}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4058, forks 522 (observed 2026-08-28T04:08:33.953013+00:00)

## What it is
Tinker Cookbook is a Python library of realistic examples and abstractions for post-training (fine-tuning) language models via the Tinker API from Thinking Machines Lab. It provides recipes for supervised fine-tuning, RL, and multimodal training while the Tinker service handles distributed training infrastructure.

## Use cases
- fine-tune a language model on custom data
- run supervised fine-tuning recipes on an LLM
- train a model with reinforcement learning post-training
- customize a language model without managing GPU infrastructure
- fine-tune a multimodal model with image and audio inputs
- experiment with post-training algorithms via an API

## When to choose
- you want to fine-tune LLMs through a managed API instead of running your own distributed training
- you need ready-made, tested recipes for SFT and RLHF-style post-training
- you want to work with Thinking Machines' Inkling models or Tinker-supported models

## When to avoid
- you need fully local, self-hosted training on your own GPUs
- you don't want to depend on a third-party API service with an account and API key
- you need frameworks like Hugging Face TRL or Axolotl for direct control over training hardware

## Facets
- artifact type: library
- maturity: active
- function: llm-training, machine-learning, sdk, agent-framework
- domain: large-language-models, machine-learning, deep-learning, developer-tools
- platform: python, cloud
- tags: fine-tuning, post-training, reinforcement-learning, supervised-fine-tuning, multimodal, api-based-training

## Member repositories
- thinking-machines-lab/tinker-cookbook (main) score 85

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:33.953013+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 2026-08-29T18:23:36.575025+00:00, confidence not recorded.
  - readme: https://github.com/thinking-machines-lab/tinker-cookbook (fetched 2026-08-28T04:08:33.953013+00:00, sha f14180f2d956)
  - registry_pypi: https://pypi.org/pypi/tinker-cookbook/json (fetched 2026-08-29T09:15:48.030669+00:00, sha 718735806e6b)
- Data as of 2026-08-30T08:39:29.467469+00:00.