What is recorded here?
Combined FP16 storage/computation, FP32 master updates and loss scaling to train tested neural networks with reduced activation bandwidth and memory while recovering higher-precision baselines.
Source-linked catalogue note. A detailed explanatory guide has not yet been prepared for this entry.
Scientific scope
Scaling must avoid overflow and gradients are unscaled before updates or clipping. FP32 updates/accumulation remain; reduced storage does not mean every operation or model trains entirely in FP16. Reported accuracy/resource outcomes are task- and implementation-specific.
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
Sharan Narang; Gregory Diamos; Erich Elsen; Paulius Micikevicius; Jonah Alben; David Garcia; Boris Ginsburg; Michael Houston; Oleksii Kuchaev; Ganesh Venkatesh; Hao Wu
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
Mixed-precision training with FP32 master weights and loss scaling
Combined FP16 storage/computation, FP32 master updates and loss scaling to train tested neural networks with reduced activation bandwidth and memory while recovering higher-precision baselines.
Scientific scope
Scaling must avoid overflow and gradients are unscaled before updates or clipping. FP32 updates/accumulation remain; reduced storage does not mean every operation or model trains entirely in FP16. Reported accuracy/resource outcomes are task- and implementation-specific.
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
Sharan Narang; Gregory Diamos; Erich Elsen; Paulius Micikevicius; Jonah Alben; David Garcia; Boris Ginsburg; Michael Houston; Oleksii Kuchaev; Ganesh Venkatesh; Hao Wu
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.