87 tools for Korean law — statutes, precedents, ordinances, interpretations | MCP Server · CLI · npm

REAP: Router-weighted Expert Activation Pruning for SMoE compression

🍀 Codebase for CloverLM

Implementation of Fast Weight Attention

Official implementation for Training LLMs with MXFP4

Implements harmful/harmless refusal removal using pure HF Transformers

Train the smallest LM you can that fits in 16MB. Best model wins!

Comparative study and experimentation on standard vs mHC vs attention residual (full and block)

A light-weight and powerful meta-prompting, context engineering and spec-driven development system for Claude Code by TÂCHES.

Original reference implementation of the CUDA rasterizer from the paper "StopThePop: Sorted Gaussian Splatting for View-Consistent Real-time Rendering"

An unofficial implementation of absGS

Open-source framework for turning expert knowledge into PII-free synthetic conversational data and production-ready LoRA adapters.

An agent for CUDA compute-communication kernel co-design

A lightweight inference engine supporting speculative speculative decoding (SSD).

Repo for FastAPI

Open-source CUDA compiler targeting multiple GPU architectures. Compiles .cu to AMD and Tenstorrent GPU's

Voice-to-text app for macOS to transcribe what you say to text almost instantly

Accelerated Asm Python

A collection of research papers on low-precision training methods

The official GitHub repo for the survey paper "A Survey on Diffusion Language Models".

Kagi Small Web

Shared Middle-Layer for Triton Compilation

Qwen3-0.6B megakernel: 527 tok/s decode on RTX 3090 (3.8x faster than PyTorch)

Polish Dataset of Banned Harmful and Offensive Content from Wykop.pl web service

Pure C inference of Mistral Voxtral Realtime 4B speech to text model

Implementation of the fast weight product key memory from Sakana AI