Tutorials

A progressive learning path from first API call to advanced training techniques. All tutorials are marimo notebooks — read here or run them interactively.

Prerequisites

Install the Tinker SDK, cookbook, and marimo:

uv pip install tinker tinker-cookbook marimo

Set your API key (get one from the Tinker Console):

export TINKER_API_KEY="your-api-key-here"

Run tutorials interactively

All tutorials are marimo notebooks in the tinker-cookbook repo. Clone the repo and run any tutorial:

git clone https://github.com/thinking-machines-lab/tinker-cookbook.git
cd tinker-cookbook
uv run marimo edit tutorials/101_hello_tinker.py

Or download a single file and run it directly:

curl -O https://raw.githubusercontent.com/thinking-machines-lab/tinker-cookbook/main/tutorials/101_hello_tinker.py
uv run marimo edit 101_hello_tinker.py

Basics

Start here. Learn the core Tinker SDK operations.

Beginner

Hello Tinker

ServiceClient, SamplingClient, and basic text generation

Beginner

Your First SFT

Renderers, Datum construction, forward-backward, optimizer step

Beginner

Async Patterns

Futures, concurrent requests, throughput optimization

Beginner

First RL

GRPO algorithm, reward functions, GSM8K math training

Core Concepts

Deep dives into specific SDK features. Each tutorial is standalone.

Intermediate

Rendering

How tokenization and chat templates work across model families

Intermediate

Loss Functions

Cross-entropy, importance sampling, PPO, custom losses

Intermediate

Completers

TokenCompleter vs MessageCompleter for RL environments

Intermediate

Weights Management

Save, load, download, and publish model weights

Intermediate

Evaluations

Evaluate your fine-tuned models during and after training

Cookbook Abstractions

Learn the higher-level training patterns from tinker-cookbook.

Intermediate

Env & EnvGroupBuilder

Core RL types: Env, EnvGroupBuilder, RLDataset, ProblemEnv

Intermediate

Custom Environment

Implementing a custom ProblemEnv subclass with format compliance

Intermediate

SFT with Config

Using train.Config and dataset builders for supervised learning

Intermediate

RL with Config

Full GRPO implementation using cookbook abstractions

Advanced

Advanced techniques for experienced users.

Advanced

SL Hyperparameters

Sweep learning rate and LoRA rank with the cookbook's sweep module

Advanced

RL Hyperparameters

KL penalty, advantage estimation, reward shaping

Advanced

DPO & Preferences

Direct preference optimization, RLHF pipeline

Advanced

Sequence Extension

Multi-turn RL with conversation history

Advanced

Multi-Agent RL

Self-play, competitive environments

Advanced

Prompt Distillation

Distilling long system prompts into model weights

Advanced

RLHF Pipeline

Full 3-stage RLHF: SFT → preference model → RL training

Deployment

Getting trained models into production.

Deployment

Export to HuggingFace

Merge LoRA adapter into a full HuggingFace model

Deployment

Build LoRA Adapter

Convert to PEFT format for efficient serving

Deployment

Publish to Hub

Upload trained models to HuggingFace Hub