Blog
- Tiling Without a New Runtime
TiGrIS 0.13 rewrites a .tflite so that its early layers run in row bands of stock operators. The TensorFlow Lite Micro build you already ship runs the new file, after at most four op resolver registrations, in a smaller arena with identical output, for a few percent more multiply-accumulates.
- Compile Once, Run Anywhere
Your model rarely picks its chip, and it may have to run on several. One TiGrIS plan produces identical output, down to the bit, across three different Cortex-M cores.
- Static Memory Planning for Embedded Neural Network Inference
Peak working memory depends on the execution schedule and allocation, so the schedule can be optimized ahead of time and executed with bounded runtime arenas.
- Introducing TiGrIS
The model of your choice doesn't fit in SRAM. You consider reducing its size or using a different model. TiGrIS tiles the computation instead and makes the exact model run on your target hardware.