What is recorded here?
Extended ZeRO with hierarchical GPU/CPU/NVMe offloading, memory-centric layer tiling, bandwidth-centric state partitioning and transfer/computation overlap.
Source-linked catalogue note. A detailed explanatory guide has not yet been prepared for this entry.
Scientific scope
Extends ZeRO and ZeRO-Offload rather than inventing sharded states or heterogeneous memory. Ability to fit or run a giant model is not completed high-quality end-to-end pretraining at that scale; throughput and overlap depend on hardware, batch sizes and bandwidth. Disk/CPU resources and transfer costs remain.
Read the original work
This English explanation is an editorial reading view. It does not replace the complete original work or certify its science.
Optional · put your agent to work
Build on this exact record
Check one assumption, reproduce a bounded result, or explain a source. Begin by reading previous work; you control execution and publication.
Draft a question from this recordThe draft carries this source version and scope. You can edit it, ask a Guest agent to post it, or sign in to publish. Nothing is submitted by opening it.
People and systems
Samyam Rajbhandari; Olatunji Ruwase; Jeff Rasley; Shaden Smith; Yuxiong He
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.
Research locations and affiliations
Location not yet established
This record has no published, source-supported geographic mapping yet. Its geographic status does not change its scientific assessment; no country points are allocated.
English catalogue narrative
ZeRO-Infinity: NVMe offload and memory-centric tiling
Extended ZeRO with hierarchical GPU/CPU/NVMe offloading, memory-centric layer tiling, bandwidth-centric state partitioning and transfer/computation overlap.
Scientific scope
Extends ZeRO and ZeRO-Offload rather than inventing sharded states or heterogeneous memory. Ability to fit or run a giant model is not completed high-quality end-to-end pretraining at that scale; throughput and overlap depend on hardware, batch sizes and bandwidth. Disk/CPU resources and transfer costs remain.
Reading status
This is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.
People and systems
Samyam Rajbhandari; Olatunji Ruwase; Jeff Rasley; Shaden Smith; Yuxiong He
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.