MuPT: A Generative Symbolic Music Pretrained Transformer
This work addresses music generation for AI and creative applications, but it is incremental as it builds on existing LLM and notation methods.
The paper tackles the problem of applying Large Language Models to symbolic music generation by proposing ABC Notation over MIDI for better compatibility and introducing Synchronized Multi-Track ABC Notation to maintain coherence across tracks, achieving models that handle up to 8192 tokens covering 90% of training data.
In this paper, we explore the application of Large Language Models (LLMs) to the pre-training of music. While the prevalent use of MIDI in music modeling is well-established, our findings suggest that LLMs are inherently more compatible with ABC Notation, which aligns more closely with their design and strengths, thereby enhancing the model's performance in musical composition. To address the challenges associated with misaligned measures from different tracks during generation, we propose the development of a Synchronized Multi-Track ABC Notation (SMT-ABC Notation), which aims to preserve coherence across multiple musical tracks. Our contributions include a series of models capable of handling up to 8192 tokens, covering 90% of the symbolic music data in our training set. Furthermore, we explore the implications of the Symbolic Music Scaling Law (SMS Law) on model performance. The results indicate a promising direction for future research in music generation, offering extensive resources for community-led research through our open-source contributions.