Build a Pipeline

Place blocks on the canvas and connect them to build a training pipeline, then save it. You can load a saved pipeline later to keep editing it.

---

Step 1. Add and connect blocks

1. In the left sidebar, click AI Training Studio → Pipeline builder.

2. Click a block in the Block library on the left, or drag it onto the canvas.

   Group	Blocks
   Data	Load dataset, Data split
   Training	SFT, DPO, CPT, GRPO
   Evaluation	Evaluation block

3. Drag a block’s output port (e.g., Dataset on Load dataset) to the next block’s input port (e.g., Train on SFT).

4. Click a block to open its settings panel on the right. A warning icon means required settings are still missing.

---

Step 2. Save

Click the save icon in the canvas toolbar, enter a Pipeline name (required) and a Description, then click Save. The canvas tab is renamed to the pipeline name.

---

Step 3. Load a saved pipeline

1. Click the folder icon at the far left of the toolbar to open the Pipelines list.

   

2. In the row’s Action (⋯) menu, choose View Pipeline. Use Delete to remove pipelines you no longer need.

3. The pipeline opens in a new canvas tab.

   

---

Next

• Run a Pipeline →