eBPF XDP on GPU

Ongoing research training transformer language models at scale, including: BERT & GPT-2

LLM inference in C/C++

Python bindings and high-level abstractions for Linux io_uring-based asynchronous I/O.

Running large language models on a single GPU for throughput-oriented scenarios.

A Datacenter Scale Distributed Inference Serving Framework

Library providing helpers for the Linux kernel io_uring support

CUDA Tile IR is an MLIR-based intermediate representation and compiler infrastructure for CUDA kernel optimization, focusing on tile-based computation patterns and optimizations targeting NVIDIA te…

[IJAIT 2021] MABWiser: Contextual Multi-Armed Bandits Library

Dynamic resources changes for multi-dimensional parallelism training

[ICML 2025 Spotlight] ShadowKV: KV Cache in Shadows for High-Throughput Long-Context LLM Inference

Accurate traffic splitting (multipath routing) technique for software switch (implemented on Open vSwitch)

🚨 Prediction of the Resource Consumption of Distributed Deep Learning Systems

eBPF implementation that runs on top of Windows

GPU-accelerated LLM Training Simulator

Userspace/GPU eBPF VM with llvm JIT/AOT compiler

AI/GPU flame graph

A Linux eBPF rootkit with a backdoor, C2, library injection, execution hijacking, persistence and stealth capabilities.

Userspace eBPF runtime for Observability, Network, GPU & General Extensions Framework

eBPF-based Security Observability and Runtime Enforcement

LMCache: Supercharge Your LLM with the Fastest KV Cache Layer

A high-throughput and memory-efficient inference and serving engine for LLMs

Linux Runtime Security and Forensics using eBPF

🗜️ Codebase-digest is your AI-friendly codebase packer and analyzer. Features 60+ coding prompts and generates structured overviews with metrics. Ideal for feeding projects to LLMs like GPT-4, Clau…

Pipeline Parallelism for PyTorch

CUDA Templates and Python DSLs for High-Performance Linear Algebra

NumPy aware dynamic Python compiler using LLVM