# Together AI Fine-Tuning Skill for Claude Code

> Skill that teaches Claude Code Together AI's fine-tuning API. Covers LoRA, full fine-tuning, DPO preference tuning, VLM training, and function-calling fine-tuning.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

```bash
npx skills add togethercomputer/skills
```

## What is This Skill?

This skill teaches AI coding agents how to fine-tune models on Together AI. It covers LoRA (efficient), full fine-tuning, DPO preference tuning, vision-language model (VLM) training, and function-calling fine-tuning — with correct API calls, data formats, and training parameters.

**Answer-Ready**: Together AI Fine-Tuning Skill for coding agents. Covers LoRA, full fine-tuning, DPO, VLM training, and function-calling tuning. Correct data formats, hyperparameters, and job management. Part of official 12-skill collection.

**Best for**: ML engineers fine-tuning open-source models on Together AI. **Works with**: Claude Code, Cursor, Codex CLI.

## What the Agent Learns

### LoRA Fine-Tuning

```python
from together import Together

client = Together()
job = client.fine_tuning.create(
    training_file="file-abc123",
    model="meta-llama/Llama-3.1-8B-Instruct",
    n_epochs=3,
    learning_rate=1e-5,
    lora=True,
    lora_r=16,
)
print(f"Job ID: {job.id}")
```

### Training Data Format (JSONL)

```json
{"messages": [{"role": "system", "content": "You are helpful."}, {"role": "user", "content": "Hi"}, {"role": "assistant", "content": "Hello!"}]}
```

### Supported Methods

| Method | Use Case | Cost |
|--------|----------|------|
| LoRA | Most tasks, efficient | Low |
| Full fine-tuning | Maximum quality | High |
| DPO | Preference alignment | Medium |
| VLM training | Vision+language | Medium |
| Function-calling | Tool use training | Low |

### Job Management

```python
# Check status
status = client.fine_tuning.retrieve(job.id)
# List jobs
jobs = client.fine_tuning.list()
# Cancel
client.fine_tuning.cancel(job.id)
```

## FAQ

**Q: Which method should I use?**
A: Start with LoRA — it is faster, cheaper, and works well for most use cases. Use full fine-tuning only if LoRA quality is insufficient.

## Source & Thanks

> Part of [togethercomputer/skills](https://github.com/togethercomputer/skills) — MIT licensed.

<!-- ZH -->

## 快速使用

```bash
npx skills add togethercomputer/skills
```

## 什么是这个 Skill？

教 AI Agent 在 Together AI 上微调模型，包括 LoRA、全量微调、DPO、VLM 训练和函数调用微调。

**一句话总结**：Together AI 微调 Skill，LoRA/全量/DPO/VLM/函数调用五种方法，数据格式和超参数，官方出品。

## 来源与致谢

> [togethercomputer/skills](https://github.com/togethercomputer/skills) — MIT

---
Source: https://tokrepo.com/en/workflows/9f339e9b-52fc-410d-af7f-f7c7779ab24f
Author: Script Depot