DeepGEMM: a clean GPU BLAS kernel library
DeepGEMM is a Cuda BLAS kernel library from DeepSeek for efficient GPU matrix math. MIT licensed, 8.7k stars, 1.3k forks.
Tag
5 posts tagged #deepseek
Browse 5 posts tagged Deepseek. Setup notes, reviews, comparisons, and workflow patterns.
DeepGEMM is a Cuda BLAS kernel library from DeepSeek for efficient GPU matrix math. MIT licensed, 8.7k stars, 1.3k forks.
ds4 is an MIT licensed C implementation of DeepSeek 4 Flash and PRO local inference, targeting Metal, CUDA and ROCm with 23,453 GitHub stars.
Whale is a DeepSeek-native terminal AI coding agent written in Go with ~98% prompt cache hit rate, 1M-token context, MCP, dynamic workflows, and Claude Code compatibility.
DeepSeek-Reasonix is a TypeScript agent framework built for DeepSeek models with structured reasoning, tool calling, and multi-step planning.
Evaluate TokenSpeed as a preview LLM inference engine for B200 benchmark reproduction, model recipes, server docs, and production-readiness checks.