Token-Equivalent Speedup
A metric for measuring an optimizer's data efficiency, expressing the percentage of tokens saved by a new method relative to a baseline when reaching the same validation loss.
Timeline (last 90 days)
On dense models, compared with MuonH, MuonGS is estimated to achieve about a 15% token-equivalent speedup when matching validation loss.
Token-equivalent speedup is a metric for measuring optimizer efficiency: the percentage of tokens saved by a new method relative to a baseline when reaching the same validation loss.
All Facts (2)
On dense models, compared with MuonH, MuonGS is estimated to achieve about a 15% token-equivalent speedup when matching validation loss.
50%UnverifiedToken-equivalent speedup is a metric for measuring optimizer efficiency: the percentage of tokens saved by a new method relative to a baseline when reaching the same validation loss.
50%