There's a 31B dense model in the Gemma 4 series that's obviously going to be sma... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

zozbot234 51 days ago | parent | context | favorite | on: Google Gemma 4 Runs Natively on iPhone with Full O...

There's a 31B dense model in the Gemma 4 series that's obviously going to be smarter (though a whole lot slower) than the MoE 26A4B.

the_pwner224 51 days ago [–]

I tried it and it was unusably slow at ~5-6 TPS. 26A4B gets close to 40 TPS which is faster than you can read, and still pretty quick with reasoning enabled.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact