@wintonARK Dojo uses bf16, so actual performance difference vs 64 bit is much less than it would seem, since you truncate so many useless zeroes & round from there
@wintonARK Dojo 使用 bf16，因此与 64 位的实际性能差异比看起来要小得多，因为您截断了这么多无用的零
@wintonARK Dojo uses bf16, so actual performance difference vs 64 bit is much less than it would seem, since you truncate so many useless zeroes & round from there
@wintonARK Dojo 使用 bf16，因此与 64 位的实际性能差异比看起来要小得多，因为您截断了这么多无用的零