Archives
All the articles I've archived.
20262
July1
Layout Matters: Optimizing AggPose Inference with OpenVINO
A practical case study on optimizing AggPose for CPU-based offline infant pose extraction with OpenVINO, INT8 quantization, pruning, distillation, and Linear-to-1x1-Conv rewriting.
March1
Exploring llama.cpp Support on OrangePi AIPro - Part 0: Enabling the CANN Backend
Part 0 of the llama.cpp on OrangePi AIPro series. Documenting the journey to enable CANN backend support on Ascend 310B, from fixing aclrtMallocHost alignment violations in RC mode to disabling unsupported NN operators. Currently running with full CPU fallback as baseline.