self-hosted/ai
§01·model · /models

YuE2-3B

musicactiveCC-BY-NC-4.0

YuE2-3B turns a style prompt and lyrics into a complete song with vocals and accompaniment, and exposes the plan it wrote as an editable ABC score, so melody, chords and lyrics can be revised and the song re-rendered. It also covers existing songs zero-shot. The vendor's quick start asks for a 24 GB NVIDIA GPU, but its own NVML figures are an 11.18 GiB peak on an RTX 4090 and 14.08 GiB at maximum context, and its pipeline caps the allocator at the card's memory minus a 2 GiB reserve - so a 16 GB card runs ordinary songs with room to spare and only the full context is out of reach, while a 12 GB card is held to a 10 GiB budget and cannot run it at all. Linux, CUDA and BF16 via a vendor Python wheel; no llama.cpp, Ollama or ComfyUI path is documented. Weights are CC BY-NC 4.0 - non-commercial use only.

§02·GPUs that run this model
11 total

benchmarked·~ runs via recipe (not benchmarked)· untested·doesn't fit