alpha=10.0
held_out=taigi_typing
model_word=(hanzi, tl_num.lower())
tokens_N=3627713
vocabulary_V=167446
records=167907
records_with_count=52126
unseen_cost=13181
cost_min=4173
cost_p10=12439
cost_p50=13181
cost_p90=13181
cost_max=13181
