DAMN for Piano Music Continuation
DAMN (Delta-Absolute Modeling with Nested Transformers) is a hierarchical autoregressive model for symbolic piano continuation, submitted to the MIREX 2026 Symbolic Music Generation task.
The model uses a score-level Transformer to model dependencies across notes and a note-level Transformer to autoregressively predict delta-onset, pitch, and duration.
Model
- Parameters: 144M
- Score-level decoder: 14 layers
- Note-level decoder: 2 layers
- Hidden size: 768
Inference
Inference code and usage instructions are available in the submission repository: naivecynics/MIREX2026-DAMN
- Downloads last month
- 5
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support