tinker-cookbook
Fine-tune, evaluate, and debug language models with ready-made training recipes
Instructs the AI agent on the repository's architecture, type system, and common pitfalls so it can correctly write, run, and debug Tinker training and evaluation code using the bundled research, debug, and inkling skills.
All Skills in This Repository (27)
Pure Emerald Level Indicators
thinking-machines-labtinker-cli
Automate Tinker training runs and checkpoint management via CLI, including HuggingFace Hub publishing.
thinking-machines-labrenderers
Converts chat-style messages into model-ready token sequences for training and inference.
thinking-machines-labevals
Evaluate language model training with inline evaluators and Inspect AI benchmarks.
thinking-machines-labci
Standardize testing conventions and GitHub Actions CI pipelines across projects.
thinking-machines-labcontributing
Provides a guide for setting up, coding, and contributing to tinker-cookbook.
thinking-machines-labenvironments
Design reusable RL environments with Env, EnvGroupBuilder, and RLDataset patterns.
thinking-machines-labweights
Downloads, merges LoRA adapters, and publishes models to HuggingFace Hub.
thinking-machines-labdatasets
Construct and normalize ML datasets across JSONL, HuggingFace, and conversation formats.
thinking-machines-labcheckpoints
Save and manage ML training checkpoints via REST API and CLI.
thinking-machines-labcompleters
Generate tokens or structured messages via SamplingClient wrappers for RL rollouts.
thinking-machines-labsetup
Guide Tinker installation, environment setup, API key provisioning, and verification.
thinking-machines-labdistillation
Automate knowledge distillation from teacher to student models via the Tinker API.
Frequently Asked Questions
FAQPage SchemaHow to install tinker-cookbook?▼
Run `npx skills add thinking-machines-lab/tinker-cookbook --all -g -y` in your terminal to install all skills in this suite globally.
What can I do with tinker-cookbook?▼
You can fine-tune language models with SFT, RL, DPO, and distillation recipes, run standard benchmarks, and export trained weights, all through the hosted Tinker training service.
Do I need my own GPUs to train models?▼
No. Your training scripts run on a regular CPU machine while the Tinker service handles all the heavy GPU work in the cloud.
Can AI agents help debug slow training runs?▼
Yes. The included debug skill walks agents through systematic triage of slow steps, hangs, renderer mismatches, and weight export issues.
Does it work with Claude Code?▼
Yes. The skills follow the universal SKILL.md standard and trigger automatically in Claude Code when you ask to set up training or diagnose issues.
Related Repositories in Software Engineering
View All in Software Engineering→
openclawopenclaw
Run a personal AI assistant across your devices and chat apps
obrasuperpowers
Gives coding agents a disciplined workflow from idea to merged code
facebookreact
AI agent skills for building, testing, and porting React core