Fine-tune LLM with Llama-Factory on Jetson

Execution model

Run one phase at a time. After each phase:

• Relay all command output to the user.
• If output contains [STOP] → stop immediately, consult the failure decision tree below.
• If output ends with [OK] → tell the user "Phase N complete" and proceed to the next phase.

Prerequisites

Phase 1 — Install jetson-examples (~2 min)

pip3 install jetson-examples
sudo reboot

[OK] after reboot completes and you can log back in.

Phase 2 — Deploy Llama-Factory (~5–15 min)

Launch Llama-Factory using the one-line deployment:

reComputer run llama-factory

This pulls the container and starts the Llama-Factory service.

Once running, open a web browser and navigate to:

http://127.0.0.1:7860

(Or replace 127.0.0.1 with the Jetson's IP for remote access.)

[OK] when the Llama-Factory WebUI loads in the browser. [STOP] if the container fails to start.

Phase 3 — Start training (~18 hours for default config)

In the WebUI:

1. Set Model name to Phi-1.5 (or your chosen model)
2. Set Dataset to alpaca_zh (or your chosen dataset)
3. Keep other training parameters as default
4. Click the Start button

Monitor training progress in the WebUI.

[OK] when training completes and the fine-tuned model appears in the save directory.

Phase 4 — Test the fine-tuned model (~2 min)

In the Llama-Factory WebUI:

1. Navigate to the Chat tab
2. Load the fine-tuned model by selecting the checkpoint path
3. Enter a prompt in the Input text box (e.g. a Chinese language prompt if using alpaca_zh)
4. Click Submit and check the output in the Chatbot text box

[OK] when the model responds with coherent output reflecting the fine-tuning data.

Failure decision tree

Reference files

• references/source.body.md — full original Seeed tutorial with screenshots, video demo, and detailed WebUI configuration (reference only)