Unified Sequence Parallelism: composing Ulysses and Ring Attention

What changed

Combined all-to-all head redistribution with ring-based sequence exchange and analyzed compatibility with tensor, data, pipeline and ZeRO parallelism for topology-aware hybrid training.

Scientific scope

Combined all-to-all head redistribution with ring-based sequence exchange and analyzed compatibility with tensor, data, pipeline and ZeRO parallelism for topology-aware hybrid training. Explicitly combines inherited Ulysses and Ring rather than inventing their components. Reported86percent MFU is for a specified LLaMA3-8B,208K-token,16-A800 setup; not a general utilization guarantee. Parallel degrees and architecture/network constraints remain.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Explicitly combines inherited Ulysses and Ring rather than inventing their components. Reported86percent MFU is for a specified LLaMA3-8B,208K-token,16-A800 setup; not a general utilization guarantee. Parallel degrees and architecture/network constraints remain.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• A Unified Sequence Parallelism Approach for Long Context Generative AI

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction