haiyewon/Decomposer-Qwen3-8B

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Decomposer-Qwen3-8B is an 8.2 billion parameter causal language model developed by haiyewon, post-trained from Qwen3-8B. This model is specifically designed to decompile symbolic music (MIDI) into executable Strudel programs. It excels at translating musical MIDI data into structured code, offering a unique capability for music-to-code generation.

Loading preview...

Decomposer-Qwen3-8B: MIDI to Strudel Program Decompiler

haiyewon/Decomposer-Qwen3-8B is an 8.2 billion parameter causal language model, post-trained from Qwen/Qwen3-8B, specializing in the decompilation of symbolic music (MIDI) into executable Strudel programs. This model is a core component of the research presented in the paper "Decomposer: Learning to Decompile Symbolic Music to Programs" (arXiv:2607.01849).

Key Capabilities

  • MIDI Decompilation: Translates MIDI input, serialized as instrument-wise note events with cycle coordinates, into Strudel program code.
  • Music-to-Code Generation: Provides a unique solution for generating programmatic representations from musical data.
  • Specialized Training: Fine-tuned on the Strudel-Synth dataset, a synthetic corpus of 21,174 (MIDI, Strudel) pairs, ensuring high accuracy in its specific domain.

Good For

  • Music Programmers: Developers and researchers working with algorithmic music composition or analysis who need to convert MIDI into a programmatic format like Strudel.
  • Creative AI Applications: Projects involving the generation or manipulation of music through code, enabling a bridge between musical input and code-based output.
  • Research in Music AI: Exploring the intersection of music, programming languages, and large language models for decompilation tasks.