CODA: Rewriting Transformer Blocks as GEMM-Epilogue Programs

Executable, measurable, and reproducible AI4AI toward recursive self-improvement. Home of OpenMLE and Frontis-MA1.

DeepStack: Facilitating Co-Design Exploration of 3D DRAM-Stacked Accelerators for Distributed LLM Inference. Includes the MICRO 2026 AE artifact.

A collection of tricks and tools to speed up transformer models

Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing…

Training a general LLM-based reasoning model for agent-compiled knowledge refinement that evolves any KB with multi-turn interaction history via RL.

Can LLMs Write Correct and Efficient GPU Communication Code?

A Micro-benchmarking Tool for HPC Networks

AI 时代的伯克希尔：基于 Claude Code / Codex 的价值投资研究框架。巴菲特·芒格·段永平·李录四大师方法论 + 多Agent并行研究。| AI-era Berkshire: a value investing research framework built for Claude Code / Codex. 4 masters' methodologies + multi…

DFlash: Block Diffusion for Flash Speculative Decoding

DeepSpec: a full-stack codebase for training and evaluating speculative decoding algorithms

a high performance library for building cache simulators

[ACL 2026] QiMeng-PRepair: Precise Code Repair via Edit-Aware Reward Optimization

Google Cloud Knowledge Catalog Tools and Samples

Debug print operator for cudagraph debugging

HRM-Text is a 1B text generation model based on the HRM architecture, strengthened by task completion and latent space reasoning.

An inference engine for HRM-Text.

Conveniently export torch.compile compiled products into self-contained Python files

An Optimizer for Nvidia Compilers.

Uni-Agent is a framework for training long-horizon agents.

Framework for evaluating and improving agents

ChampSim is an open-source trace based simulator maintained at Texas A&M University and through the support of the computer architecture community.

DeepSeek 4 Flash and PRO local inference engine for Metal, CUDA and ROCm

cuda-oxide is an experimental Rust-to-CUDA compiler that lets you write (SIMT) GPU kernels in safe(ish), idiomatic Rust. It compiles standard Rust code directly to PTX — no DSLs, no foreign languag…

A native macOS app for syncing projects to your local filesystem

👾 Open Computer Use – Open-Source Alternative to Codex Computer Use

Compact and Agent-Native MoE Training System

high-performance linear attention kernel library built on TileLang