Transferring Extreme Subword Style Using Ngram Model-Based Logit Scaling
This addresses stylistic text generation for applications like creative writing or character simulation, but appears incremental as it builds on existing logit scaling methods.
The paper tackles the problem of transferring extreme subword stylistic variation to large language models at inference time using an ngram model-based logit scaling technique, with results evaluated by tracking perplexity metrics to balance adaptation and fluency.
We present an ngram model-based logit scaling technique that effectively transfers extreme subword stylistic variation to large language models at inference time. We demonstrate its efficacy by tracking the perplexity of generated text with respect to the ngram interpolated and original versions of an evaluation model. Minimizing the former measure while the latter approaches the perplexity of a text produced by a target author or character lets us select a sufficient degree of adaptation while retaining fluency.