Pinned Loading

1. Sigmoid-TopK-Fusion Sigmoid-TopK-Fusion Public

   Fused Sigmoid+TopK Triton kernel for MoE routing — 3.1x faster than PyTorch baseline. Inspired by Sarvam AI's sovereign model inference stack.

   Jupyter Notebook 2

2. High-Performance-Reduction-Kernels High-Performance-Reduction-Kernels Public

   CUDA C reduction kernels benchmarking with Triton, PyTorch and CUB primitives

   Jupyter Notebook 5 1

3. mossformer2-denoise mossformer2-denoise Public

   Production-ready MossFormer2_SE_48K speech denoising — reference Python (clearvoice + PyTorch) and lean Rust (ONNX ▎ Runtime) Docker containers with matching CLIs. Includes reproducible ONNX expor…

   Jupyter Notebook

4. hexkernels hexkernels Public

   538 Hexagon NSP kernels that actually use the accelerator (HVX/HMX), plus the agentic pipeline, simulator/silicon measurement, and the ELF-disassembly anti-cheat detector that proves it

   C