Throughput-oriented systems for large language models on commodity GPUs. - View it on GitHub