Skip to main content

Why Nikla's MoE w new tokenizer spikey val?

Created on August 1|Last edited on August 1

Section 1


5k10k15k20k25k30kStep20406080100
5k10k15k20k25k30kStep123
5k10k15k20k25k30kStep46810
5k10k15k20k25k30kStep22.533.54
5k10k15k20k25k30kStep20406080
Run: olmoe-8x1b-newhp-newds-cx5-fine1-newtok
1
Run set 2
1
Run set 3
Run set 4