Skip to main content

Using Fine-Tuning and Status Monitoring

Once the fine-tuning instance launches successfully, click on the instance name to enter the operation console, configure parameters, and bind datasets.

Using Web Interface​

If you selected a framework with a visual interface (like LLaMA-Factory) during creation, the system overview page directly presents the tuning console.

Fine-tune Overview

From this screen, you can:

  1. Load training stage configurations.
  2. Select dataset.
  3. Adjust training hyperparameters (e.g., Batch Size, Learning Rate, Epochs).
  4. Click [Start Fine-Tuning] at the bottom.

Using Notebook Interactive Environment​

Users requiring highly customized workflows can click the [Start Notebook] button.

Start Notebook

This opens a JupyterLab development environment containing deep system access, perfect for making advanced script modifications and executing validation operations.

Training Status and Analysis​

Upon starting fine-tuning, you can view real-time telemetry indicators by navigating to the [Analysis] tab sequentially in the console side menu:

Analysis Status Data

  • Monitor CPU/GPU utilization across instances to assess computational load.
  • Check current memory and VRAM allocation.
  • Display relevant metric plots and training curves (e.g., Loss tracking).

Billing and Settlement​

Resource consumption statistics and precise bills can be accessed via the [Billing] tab in the sidebar:

Fine-tune Instance Billing

Note: Please remember to proactively stop or delete your instance in the Settings panel when all tasks are complete down to deter continuous billing costs.