How Far Can Pretrained LLMs Go in Symbolic Music? Controlled Comparisons of Supervised and Preference-based Adaptation
This work addresses the effectiveness of LLM adaptation for symbolic music applications, providing insights for researchers and practitioners in music AI, though it is incremental in nature.
The study compared finetuning strategies for adapting pretrained large language models to symbolic music tasks, finding trade-offs between domain adaptation and preserving prior knowledge, with varied metric behaviors.
Music often shares notable parallels with language, motivating the use of pretrained large language models (LLMs) for symbolic music understanding and generation. Despite growing interest, the practical effectiveness of adapting instruction-tuned LLMs to symbolic music remains insufficiently characterized. We present a controlled comparative study of finetuning strategies for ABC-based generation and understanding, comparing an off-the-shelf instruction-tuned backbone to domain-adapted variants and a music-specialized LLM baseline. Across multiple symbolic music corpora and evaluation signals, we provide some insights into adaptation choices for symbolic music applications. We highlight the domain adaptation vs.~preserving prior information tradeoff as well as the distinct behaviour of metrics used to measure the domain adaptation for symbolic music.