Ultra and Unified CCL

OpenArm CAD data and information for manufacturing

A fast inference library for running LLMs locally on modern consumer-class GPUs

🤗 PEFT: State-of-the-art Parameter-Efficient Fine-Tuning.

Nano vLLM

ROCm Communication Collectives Library (RCCL)

Optimized primitives for collective multi-GPU communication

Development repository for the Triton language and compiler

NEO is a LLM inference engine built to save the GPU memory crisis by CPU offloading

A tiny yet powerful LLM inference system tailored for researching purpose. vLLM-equivalent performance with only 2k lines of code (2% of vLLM).

Multi-LoRA inference server that scales to 1000s of fine-tuned LLMs

GMT: GPU Orchestrated Memory Tiering for the Big Data Era

Kmesh website and documentation repo

High Performance ServiceMesh Data Plane Based on Programmable Kernel