Acyr Locatelli

Papers that credit Acyr Locatelli in their source bylines. Matching names may refer to different researchers.

2 results

1. PaperNLP & Language

   Synthetic training fails to beat text on fixed budgets

   

   Tests of synthetic pre-pretraining find no fixed-budget language modeling advantage over natural-language-only training in the settings studied.

   Added to Chiosa Oct 30, 2026
2. PaperMachine Learning

   Why sparse MoE models gain at low batch sizes

   

   Measurements show why sparse MoE models can gain at low batch sizes, while an IPF framework estimates expert reuse in tested models.

   Added to Chiosa Oct 30, 2026