Note
Sentis is now called Inference Engine. The documentation has moved to https://docs.unity3d.com/Packages/com.unity.ai.inference@latest. Refer to the new location for the latest updates and guidance. Make sure to update your bookmarks and references accordingly.
Profile a model
The performance of a model depends on the following factors:
- The complexity of the model.
- Whether the model uses performance-heavy operators such as
ConvorMatMul. - The features of the platform you run the model on, for example CPU memory, GPU memory, and number of cores.
- Whether Sentis downloads data to CPU memory when you access a tensor. Refer to Get output from a model for more information.
Profile a model in the Profiler window
To get performance information when you run a model, you can use the following:
- The Profiler window.
- RenderDoc, a third-party graphics debugger.
The Profiler window displays each Sentis layer as a dropdown item in the Module Details panel. Open a layer to get a detailed timeline of the execution of the layer.
When a layer executes methods that include Download or Upload, Sentis transfers data to or from the CPU or the GPU. This might slow down the model.
If your model runs slower than you expect, refer to:
- Understand models in Sentis for information about how the complexity of a model might affect performance.
- Create an engine to run a model for information about different types of worker.