1d ago

Sasha Rush says the base architecture of models under discussion matches Composer 2, the neural network training library developed by MosaicML

Exchange formed part of ongoing technical conversation on scaling.

——0——
Original post
OPsamsja#1262samsja@SAMSJA19

@srush_nlp @eliebakouch did you guys did dsa 🫢

12:47 PM · May 18, 2026 View on X
REPLYSasha Rush#23Sasha Rush@SRUSH_NLP

@samsja19 @eliebakouch Basically same base arch as Composer 2.

samsjasamsja@samsja19

@srush_nlp @eliebakouch did you guys did dsa 🫢

7:47 PM · May 18, 2026 · 27 Views
7:55 PM · May 18, 2026 · 70 Views