🧠LLM Fine-Tuner — Documentation¶
Welcome! This documentation covers everything you need to go from zero to a fine-tuned, deployed AI model — no coding required.
What is LLM Fine-Tuner?¶
LLM Fine-Tuner is a graphical tool that lets you take a pre-trained AI language model and teach it new skills using your own data. Think of it like taking a general-purpose assistant and training it to become an expert in your specific domain — whether that's customer support, writing in a particular style, answering medical questions, or anything else.
You upload your data, click a button, and get back a custom model ready to use.
Who is this for?¶
- Non-technical users — if you can use a spreadsheet, you can use this tool
- Researchers — full control over every training parameter
- Small teams — no GPU cluster needed; runs on a single gaming GPU
- Developers — full CLI for automation and scripting
Documentation Map¶
| Guide | What you'll learn |
|---|---|
| 01 — Installation | How to install on Windows, Mac, Linux, or Google Colab |
| 02 — Quick Start | Train your first model in 5 minutes |
| 03 — Data Preparation | How to format your data, upload it, and fix column issues |
| 04 — Training | All training modes, presets, and settings explained |
| 05 — RLHF Pipeline | Reward models, GRPO, KTO and ORPO alignment |
| 06 — Inference | Testing your model, batch runs, and vLLM serving |
| 07 — Evaluation | Measuring quality with BLEU, ROUGE, BERTScore |
| 08 — Export & Deploy | Download, push to HuggingFace Hub, export to GGUF |
| 09 — CLI Reference | Full command-line interface with examples |
| 10 — Advanced Usage | Heretic Mode, Flash Attention, QLoRA, hardware tips |
| 11 — Troubleshooting | Common errors and how to fix them |
| 12 — FAQ | Frequently asked questions |
System Requirements¶
| Component | Minimum | Recommended |
|---|---|---|
| Python | 3.10 | 3.11 or 3.12 |
| RAM | 8 GB | 16 GB+ |
| GPU VRAM | None (CPU only) | 8 GB+ (NVIDIA) |
| Disk space | 10 GB | 50 GB+ |
| OS | Windows 10, macOS 12, Ubuntu 20.04 | Any modern OS |
No GPU? You can still fine-tune small models (gpt2, distilgpt2) on CPU. It will be slow but it works.