LoongServe GitHub repository (elastic sequence parallelism, ASP 2024)

README is three lines: 'This is an implementation of the paper: LoongServe: Efficiently Serving Long-context Large Language Models with Elastic Sequence Parallelism. To reproduce all the main results in our paper, please check the artifact folder and follow the instructions in it.' Repository language is Jupyter Notebook; contents are docs, longserve_c_scheduler, longserve_cuda_kernels, loongserve, requirements.txt, rnccl, run_multi_nodes.sh, setup.py, test. No GitHub releases. The docs/artifact-eval/README.md artifact-evaluation guide is written explicitly for AE (Artifact Evaluation) reviewers: it walks through logging into the AUTHORS' CLUSTER via a jump server, activating the environment, and reserving compute nodes - with a separate 'Instructions for non-AE-reviewers' section. Baseline comparisons scripted in the AE guide include vLLM, LightLLM with SplitFuse, DeepSpeed-MII with Dynamic SplitFuse, and DistServe with prefill-decode disaggregation, across ShareGPT, LEval, LV-Eval and Mixed workloads. API metadata 2026-10-04: created 2024-08-21, last push 2024-11-11, 137 stars, 16 forks, 5 open issues, Apache-2.0, not archived.