Fine-tune with LoRA

How to create a fine-tuning job with LoRA?

To fine-tune a model with LoRA, follow the instructions below.

note

• You must log in before starting a fine-tune job.
• Ensure you have enough balance (credit).
• At least one base model must be available for fine-tuning.

Steps

1. Go to the Fine-tuning Jobs page and click + Fine-tune model.
2. In the pop-up, enter the Name of your fine-tuning job.
   • Validation: required, max 100 characters, supports Unicode letters, digits, -, _, .

3. Select a base model from the dropdown list.
   • Examples: gemma-3-27b-it, Qwen3-4B-Instruct-2507, Llama-3.3-70B-Instruct
4. Select dataset format from the dropdown list: Alpaca / ShareGPT / ShareGPT_Image.
5. Upload your training data file.
   • Supported formats: CSV, JSON, JSONL, ZIP, Parquet (< 100 MB).
6. (Optional) Upload validation data.

7. (Optional) Configure hyperparameters:
   • Learning rate: float, 1e-6 → 1e-4 (e.g., 0.00001)
   • Number of epochs: integer 1–20 (default = 5)
8. Click Create to start the fine-tuning job.
   • The job will appear in the table with status Running.

tip

Fine-tuning with LoRA usually takes only a few minutes.

---

How to manage fine-tuning jobs?

On the Fine-tuning Jobs page, you can:

• View detail — open the pipeline detail in AI Studio.
• Deploy model — once training is completed, deploy the LoRA model.
• Cancel job — cancel a running job (requires confirmation).
• Delete job — permanently delete a job (requires confirmation).

Status badges

• Running (yellow)
• Succeeded (green)
• Failed (red)
• Canceled (gray)