Machine Learning Engineer obsessed with ML Systems, Inference Optimization, and PyTorch internals. Here I am documenting everything I make or work on.. soo yeah.
Identity Is Not the Keyword
Adversarial disentanglement (GRL + CLUB), dual-gate scoring, and four-phase training produce a 0.60 MB model for custom wake word detection. Achieves 4.69% keyword EER and 0.8425 joint AUC with only 1.8M parameters — deployable on CPU at 26 ms per 2-second window.
Deconstructing my PyTorch ATen PR: What happens when you index a Tensor?
How I fixed a crash in PyTorch’s ATen C++ layer when tensor indexing receives an empty index list — tracing the dispatch from Python through to IndexingUtils.h and TensorAdvancedIndexing.cpp.