Stars
TurboQuant: Near-optimal KV cache quantization for LLM inference (3-bit keys, 2-bit values) with Triton kernels + vLLM integration
adds /goal functionality similar to that used in Codex and Claude Code to OpenCode.
Ideogram 4: Open image model at the forefront of design
NVIDIA FastGen: Fast Generation from Diffusion Models
Conveniently export torch.compile compiled products into self-contained Python files
TokenSpeed is a speed-of-light LLM inference engine.
Ready-to-use ML training recipes to help you build and deploy models on Baseten.
The lightweight framework for building agents
A fast type checker and language server for Python
Performant kernels, and other ML Systems integrations
From a+b to sparsemax(QK^T)V in Triton!
Anthropic's original performance take-home, now open for you to try!
Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels
A kernel library written in tilelang
CUDA kernels for linear attention variants, written in CuTe DSL and CUTLASS C++.
Use Codex from Claude Code to review code or delegate tasks.
Region-level profiling for CUDA kernels with trace, NVBit, CUPTI, NSys, and an interactive Explorer.
Dashboard for InferenceX™, Open Source Continuous Inference
Skills for Real Engineers. Straight from my .claude directory.
Train speculative decoding models effortlessly and port them smoothly to SGLang serving.
Agent Lattice: a knowledge graph for your codebase, written in markdown.
A markdown native slides tool for academics building with agents.
CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies
Gemini auth plugin for opencode