# tinker

Fine-tune LLMs with the Tinker API using SFT, LoRA, and RL across text, vision, and audio models like Inkling.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Windsurf (Devin Desktop), Claude.ai · 1 install · ★ 152 repo stars
- Author: sundial-org
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/sundial-org/skills/tree/HEAD/skills/tinker/SKILL.md
- Updated: 2026-07-15
- Security check: automated, 2026-11-16 (text) — no signs of malicious code found.
- HTML: https://affpapa.org/en/nevibecode/skills/tinker

## What it does

This skill teaches your coding agent how to build fine-tuning workflows against the Thinking Machines Lab API. It handles high-level recipes via `tinker-cookbook` for automated data preparation, as well as bare primitives like `forward_backward()` and `optim_step()` when you need full control over the training loop.

The reference covers diverse model families. You can fine-tune Qwen3.5/3.6, Kimi K2.x, or the 975B MoE model Inkling (41B active) with support for image, audio, and tunable thinking effort. For reinforcement learning, it includes patterns for PPO and GRPO-style grouped rollouts.

To run actual jobs, you will need the Python packages `tinker` and `tinker-cookbook`. Without setting your credentials in `TINKER_API_KEY`, API requests will fail immediately.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add sundial-org/skills --skill tinker -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/tinker/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/tinker/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add sundial-org/skills --skill tinker -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/tinker/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add sundial-org/skills --skill tinker -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/tinker/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install sundial-org/skills skills/tinker
```

2. **Manually: put the files in place**
- File: `.github/skills/tinker/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add sundial-org/skills --skill tinker -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/sundial-org/skills --path skills/tinker
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/tinker/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add sundial-org/skills --skill tinker -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/sundial-org/skills/tree/HEAD/skills/tinker/SKILL.md

## When to use it
- configure LoRA fine-tuning for Qwen or Kimi models
- build supervised fine-tuning pipelines using chat dataset builders
- implement reinforcement learning with PPO or GRPO loss functions
- train multimodal Inkling models using visual and audio inputs

## Example prompts
- Write a low-level Tinker script to train a LoRA adapter on Qwen3.6-35B-A3B
- Set up an SFT pipeline using tinker-cookbook for a Hugging Face dataset
- Show how to configure thinking effort when fine-tuning Inkling

## Similar
- [tinker-ai-skill](https://affpapa.org/en/nevibecode/skills/tinker-ai-skill.md) — skill · ★ 330 repo stars
- [fine-tuning-expert](https://affpapa.org/en/nevibecode/skills/fine-tuning-expert.md) — skill · 4600 installs
- [axolotl](https://affpapa.org/en/nevibecode/skills/axolotl.md) — skill · 1100 installs
- [llm](https://affpapa.org/en/nevibecode/skills/mindrally-llm.md) — skill · 767 installs
- [llm-fine-tuning-guide](https://affpapa.org/en/nevibecode/skills/llm-fine-tuning-guide.md) — skill · 196 installs
- [transformers-huggingface](https://affpapa.org/en/nevibecode/skills/transformers-huggingface.md) — skill · 802 installs

From the same set: [sundial-org/skills](https://affpapa.org/en/nevibecode/by/sundial-org/skills.md) — 2 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md
