Pinned Loading

1. retriforever retriforever Public

   KISS! A local retrieval layer for AI code assistants. No cloud, no embedding magic, no subscription for what ctags used to do.

   Python

2. mlops mlops Public

   CUDA training ops for GPUs without native FP4 (sm_50..sm_90): AdamW and in-place NVFP4 AdamW-QAT, C API + PyTorch, FBLA Attention... and more (will be)

   Python 1

3. nvfp4effma nvfp4effma Public

   NVFP4 GEMM, GEMV, linear layers and quantization for NVIDIA GPUs without native FP4 (sm_50–sm_90): every NVIDIA NVFP4 variant (ModelOpt, Transformer Engine 1x16 / 16x16, 4over6, swizzled scales) bi…

   Cuda

4. e5m2effma e5m2effma Public

   FP8 (E5M2 and E4M3, any pairing) GEMM, GEMV and fused AdamW for NVIDIA GPUs without native FP8 (sm_50–sm_87, Maxwell to Ampere): exact products, FP32 accumulation, TE HYBRID backward, C API and PyT…

   Cuda

5. e4m3effma e4m3effma Public

   FP8 (E4M3) GEMM, GEMV and fused AdamW for NVIDIA GPUs without native FP8 (sm_50–sm_87, Maxwell to Ampere): exact products, FP32 accumulation, C API and PyTorch bindings incl. torch._scaled_mm and F…

   Cuda 1

6. x.AGENTS.md x.AGENTS.md Public

   A comprehensive coding style and operational protocol for AI agents — mandatory rules, best practices, and language-specific guidance to produce maintainable, secure, and reliable code.

   2