Search Results
7/9/2025, 7:04:41 PM
>>105848408
>How long did the training take?
To my taste, ~8hrs, ~3.5k steps at 640 for bs=3 and float8 transformer. Maybe it needs more, and maybe it needs less.
>How long did the training take?
To my taste, ~8hrs, ~3.5k steps at 640 for bs=3 and float8 transformer. Maybe it needs more, and maybe it needs less.
Page 1