AWhat changed
Integrated multi-GPU execution and architecture-aware inference with CPU/NVMe offloading to support dense and expert Transformer models under different latency and throughput constraints.
Scientific scope
Integrated multi-GPU execution and architecture-aware inference with CPU/NVMe offloading to support dense and expert Transformer models under different latency and throughput constraints. Performance and model-capacity figures depend on hardware, workloads and resource budgets. Distributed trillion-parameter inference is not single-GPU capability. Original training-side ZeRO and later FastGen remain separate.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Performance and model-capacity figures depend on hardware, workloads and resource budgets. Distributed trillion-parameter inference is not single-GPU capability. Original training-side ZeRO and later FastGen remain separate.
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.
- DeepSpeed Inference: Enabling Efficient Inference of Transformer Models at Unprecedented Scale
- Aminabadi et al.: DeepSpeed dense, sparse and heterogeneous inference — official research record
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