3 ms·

My burning question: Why not also make a slightly larger model (100B) that could perform even better? Is there some bottleneck there that prevents RL from scal

My burning question: Why not also make a slightly larger model (100B) that could perform even better? Is there some bottleneck there that prevents RL from scaling up performance to larger non-MoE model?

t1amat 2y ago
See QwQ-Max-Preview: https://qwenlm.github.io/blog/qwq-max-preview/ https://qwenlm.github.io/blog/qwq-max-preview/
buyucu 2y ago
they have a larger model that is in previes and still training.