Can Perplexity Predict Fine-Tuning Performance? An Investigation of Tokenization Effects on Sequential Language Models for Nepali

arXiv:2404.18071v1 Announce Type: cross
Abstract: Recent language models use subwording mechanisms to handle Out-of-Vocabulary(OOV) words seen during test time and, their generation capacity is generally measured using perplexity, an intrinsic metric. It is known that increasing the subword granularity results in a decrease of perplexity value. However, the study of how subwording affects the understanding capacity of language models has been very few and only limited to a handful of languages. To reduce this gap we used 6 different tokenization …

abstract arxiv capacity cs.cl cs.lg effects fine-tuning intrinsic investigation language language models performance perplexity results test tokenization type words