trends.barnas.net

method · first seen 4 Oct 2026

chunked loss computation

Computes a large loss over a big output vocabulary in chunks so the full logits tensor never sits in memory.

Mentions per week

0119 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Earliest sources

Latest sources

Mentioned together with

sequential recommendationbinary cross-entropy lossfused kernel for loss and gradient