Zobrazeno 1 - 1
of 1
pro vyhledávání: '"Manevich, Avashalom"'
Autor:
Lieber, Opher, Lenz, Barak, Bata, Hofit, Cohen, Gal, Osin, Jhonathan, Dalmedigos, Itay, Safahi, Erez, Meirom, Shaked, Belinkov, Yonatan, Shalev-Shwartz, Shai, Abend, Omri, Alon, Raz, Asida, Tomer, Bergman, Amir, Glozman, Roman, Gokhman, Michael, Manevich, Avashalom, Ratner, Nir, Rozen, Noam, Shwartz, Erez, Zusman, Mor, Shoham, Yoav
We present Jamba, a new base large language model based on a novel hybrid Transformer-Mamba mixture-of-experts (MoE) architecture. Specifically, Jamba interleaves blocks of Transformer and Mamba layers, enjoying the benefits of both model families. M
Externí odkaz:
http://arxiv.org/abs/2403.19887