What is TT-NN?
TT-NN is an open source library of neural network operations built on the tt-metal programming model.
It can be used in C++ and Python projects, and is designed to feel familiar to developers experienced with PyTorch.
Key features of TT-NN include:
More than 200 operations (matrix multiplication, convolution, reduction, CCL, fused Transformer operations, etc.)
A Tensor type that enables different ways to represent, distribute and access data on device
The ability for developers to register custom operations
Native support for a mesh of devices
Utilities to cache converted parameters, significantly speeding up model loading on repeated runs
A comparison mode that enables debugging of long operation sequences against a known reference