Create a project

Prepare the task

Write down the scene, what the model should recognize, and how you will judge a useful result. For example: identify pallets and forklifts in warehouse images. Keep unrelated tasks in separate projects so their classes and evaluation data remain clear.

Create and name the project

1. Open the training-project area and choose the create action.
2. Enter a descriptive name and a short description of the scene and data source.
3. Choose the project type offered by your deployment. CV object detection uses images and object labels. VLM work uses images with question-and-answer or instruction data.
4. Save, then open the project to check that its type and name are correct.

Some interfaces display Chinese labels such as 训练项目 (training projects) or 创建项目 (create project). The guide translates their meaning; button wording can vary with the installed version.

Check the project before continuing

Confirm that you can open its data and training areas. If no data is present, import a small dataset first. If a project cannot be loaded, use its retry control and check the account you signed in with.

Pick the right type

A CV dataset and a VLM instruction dataset are different inputs. Do not rename a file to change its format. Native text-only LLM workflows are outside this guide’s verified scope. Ask the team before preparing a large import for an unsupported type.

Next: import data.