A large-scale vector HD-map dataset for autonomous driving research.

A geometry-based visual sampler compressing 2D images into 1D signals for high-efficiency perception.

A memory-efficient optimizer that reduces AdamW memory footprint via second-moment sharing and quantization.

An MIT-licensed framework for automatically upgrading AI agent skills via performance-based optimization.

An Adaptable text Representations Compressor for reducing LLM inference costs.

Open-source agentic harness for LLM-based generation and optimization of CUDA kernels.

Open-source unified multimodal understanding and generation model family.

KV cache compression using Retrieval-reasoning and Logic-constructed Attention Behaviors

Agentic and interactive theorem proving system with integrated Lean 4 verification.

Open-access agentic AI platform for materials design on AtomGPT.org.

Speculative decoding framework using semi-autoregressive generation to accelerate LLM inference.

A lightweight optimizer for embedding and LM-head matrices with O(V + D) VRAM complexity.

A native Metal inference runtime for LLMs on Apple Silicon designed for high throughput.

Content-Aware Recurrent with Value Efficiency for chunk-parallel linear attention.

AI-assisted system for discovering cryptographic misuse vulnerabilities in libraries and applications.

Open-source toolkit for building and evaluating biomedical deep research agents.

Open-source hardware and software for a physical reinforcement learning robot platform.

Automated theorem proving tool for formally verifying computer systems using LLMs.

A block-sparse attention kernel that reduces attention compute by 28.4x at 1M context.

A self-play RL method for open-ended LLM self-improvement using verifiable proxy environments.