tinker-cookbook is a Python library by Thinking Machines Lab providing abstractions for fine-tuning LLMs through the Tinker API, which offloads distributed training to their infrastructure.

It includes recipes for SFT, RL (math, code, multi-agent), DPO/RLHF, distillation, tool use, audio, and VLM tasks, alongside 20+ marimo notebooks. The library supports their Inkling and Inkling-Small models with controllable reasoning effort and multimodal audio/image inputs, and is installable from PyPI.

Topics

Reinforcement Learning Tinker LoRA Post-Training LLM Fine-Tuning

View on GitHub All Bookmarks

Discover Similar Content

William's Bookmark Library

/**/

thinking-machines-lab/tinker-cookbook: Post-training with Tinker

github.comSaved August 16, 20265 min

thinking-machines-lab · via GitHub

Summary

tinker-cookbook is a Python library by Thinking Machines Lab providing abstractions for fine-tuning LLMs through the Tinker API, which offloads distributed training to their infrastructure.

Topics

Reinforcement Learning Tinker LoRA Post-Training LLM Fine-Tuning

View on GitHub All Bookmarks