List your company

Character level language modeling on enwik8 (dev)

1BPCAdaptive
0.99761.01381.031.0462May 19, 2019Feb 16, 2020Nov 15, 2020Aug 16, 2021May 16, 2022Feb 13, 2023Nov 14, 2023

Evaluation results

MethodLinks
Adaptive#layers=24, #Params=209M, #FLOPS=181M, Span limit S=81922019.051
Transformer-LS#Param=44M2021.071.01
Longformer#Param=41M2020.041.02
Longformer#Param=41M2021.071.02
Adaptive#layers=12, #Params=39M, #FLOPS=41M, Span limit S=81922019.051.04
Adaptive#Param=39M2020.041.04
Adaptive#Param=38M2021.071.04
Adaptive Attention SpanContext=8192, #Params=39M2023.111.04
Skip Cross-Head Transformer-XLContext=3800, #Params=41M2023.111.058
Skip Cross-Head Transformer-XLContext=1600, #Params=41M2023.111.06