Fine‑Tune Large AI Models on Huawei Cloud in One Minute

Fine‑tuning (Fine Tuning) is the key to using large models cost‑effectively, allowing them to answer domain‑specific questions that a generic model cannot handle.

For example, a general model can converse fluently, but to correctly answer "Can ibuprofen be taken together with traditional Chinese medicine?" we must fine‑tune it with relevant data.

Pre‑trained large models act like general‑purpose talent, capable of answering questions, summarizing data, writing code, etc., yet no single model solves every specialized problem. Fine‑tuning with domain‑specific datasets transforms them into experts for fields such as law, finance, medicine, or art.

Step 1: Choose a Model

Visit the AI Gallery website, click “Model” to view the list of models adapted for Huawei Ascend Cloud. Select a suitable open‑source model (e.g., GLM/ChatGLM).

Step 2: Select a Pre‑set Dataset and Start Fine‑Tuning

On the model detail page, go to “Train > Fine‑Tuning Master” to open the fine‑tuning workflow. Upload the dataset that matches your fine‑tuning goal (e.g., an advertising‑copy optimization dataset) and configure related parameters.

Choose the recommended compute specification for the selected model; the platform provides optimal configurations based on extensive developer experience.

After the training finishes, you can view training metrics such as resource utilization and loss curves.

Step 3: Deploy the Model as an Online Inference Service

In the deployment page, configure service parameters.

Click “Start Deployment”.

Confirm service details and cost on the order confirmation page, then submit.

When the status changes to “Running”, the service is ready and the API can be called directly.

Step 4: Verify Model Effectiveness

Select the fine‑tuned model from the fine‑tuning list, deploy it as an AI application, and test its responses directly to ensure the desired improvement.