TT-LLK
Low-Level Kernels (LLKs) enable AI models to operate on Tensix cores efficiently.
Use our LLKs to efficiently build complex functions and run AI operations on Tenstorrent hardware.
Purpose and Functionality
LLKs control the Tensix Engine, the fundamental computational unit of the Tensix Core. LLKs also provide an easy way to use the Tensix instruction set.
LLKs enable low-level mathematical operations to be executed at peak performance.
LLKs and the Tensix Core
LLKs provide developers with a direct, optimized pathway to program the hardware, ensuring that every operation is executed with minimal overhead and maximum efficiency.
A compiler that bridges machine learning frameworks with Tenstorrent hardware, using LLKs to optimize the execution of high-level models on Tensix cores.
An open-source SDK that facilitates the development of custom kernels, offering low-level access to Tenstorrent’s hardware and leveraging LLKs for optimized performance.
Development and Testing
To get started, check out the LLK repository on GitHub. Developers can access an independent CI pipeline and testing environment and build on Wormhole and Blackhole.