How ckorikov/2025-ttie works
Project type and purpose
• This is a C++-based library and accompanying example/executable for “Toy Training and Inference Engine (TTIE)” that implements neural-network training and inference on various back-ends (CPU, GPU, NPU). It targets efficient AI model execution.
File structure
• Top level:
• CMakeLists.txt – project build orchestration
• /ttie/ttie.h – public API header for the library
• example/main.cpp (+ run.sh) – sample client application using the library
• tests/test_main.cpp – unit-test driver
• .github/workflows/*.yml – CI build & test definitions
• No separate “src/” folder; library sources are likely under include/ttie or implicitly added via CMake.
• Configuration: .clang-format, .clang-tidy, CMake-based build.
README for additional insights
• The library defines Layers (convolution, pooling, normalization, activations, RNNs, etc.), Loss and Distance functions, Tensor abstraction, and multiple hardware back-ends.
• Emphasis on performance optimizations (parallelism, operator fusion, quantization).
Final recommendations
• Core component (Tensor, Layer, Backend) as a separate node.