##############################################
step: final
Table for all datasets except crossNER
ACE 2004            : 26.0% 36.9%
ACE 2005            : 23.6% 37.0%
AnatEM              : 30.6% 35.8%
Broad Tweet Corpus  : 60.2% 65.0%
CoNLL 2003          : 63.6% 68.5%
FabNER              : 24.0% 34.5%
FindVehicle         : 44.1% 44.6%
GENIA_NER           : 53.6% 58.9%
HarveyNER           : 24.9% 22.6%
MultiNERD           : 63.9% 71.0%
Ontonotes           : 31.6% 45.6%
PolyglotNER         : 42.8% 50.3%
TweetNER7           : 40.1% 51.6%
WikiANN en          : 58.1% 67.8%
WikiNeural          : 72.3% 77.1%
bc2gm               : 52.3% 59.0%
bc4chemd            : 51.4% 56.2%
bc5cdr              : 69.4% 73.5%
ncbi                : 61.1% 70.2%
Average             : 47.0% 54.0%

Table for zero-shot benchmark
CrossNER_AI         : 60.1% 64.7%
CrossNER_literature : 64.9% 72.5%
CrossNER_music      : 69.9% 72.9%
CrossNER_politics   : 72.7% 79.5%
CrossNER_science    : 66.4% 72.0%
mit-movie           : 55.1% 63.4%
mit-restaurant      : 44.6% 57.0%
Average             : 62.0% 68.9%
##############################################