Yu et al.: Orca iteration-level generative serving

What changed

Scheduled autoregressive inference one iteration at a time and selectively batched operations, enabling finished requests to leave and new requests to enter active generation batches.

Scientific scope

Scheduled autoregressive inference one iteration at a time and selectively batched operations, enabling finished requests to leave and new requests to enter active generation batches. Reported throughput/latency gains depend on the comparison engine, batching and hardware workload. This is scheduling, not the later PagedAttention KV-memory mechanism; no universal speedup is implied.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Reported throughput/latency gains depend on the comparison engine, batching and hardware workload. This is scheduling, not the later PagedAttention KV-memory mechanism; no universal speedup is implied.

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.

• Orca: A Distributed Serving System for Transformer-Based Generative Models

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