YuE2-Vae merge 0.666

A drop-in decoder for YuE2: a linear, per-tensor weight merge of YuE2-Vae (weight 0.666) and YuE2-Vae-legacy (weight 0.334).

The two releases share an identical architecture and an identical encoder; only the decoder differs, and the weights sit ~5% apart (fine-tune lineage), so interpolation is well-behaved. 0.666 was chosen for more than one reason.

from yue2 import YuE2Pipeline
pipe = YuE2Pipeline.from_pretrained("m-a-p/YuE2-3B", vae="Mothersuperior/YuE2-Vae-merge-0.666")

Weights are CC BY-NC 4.0, inherited from the source checkpoints (see LICENSE). Not affiliated with m-a-p.

Downloads last month: 252

Safetensors

Model size

0.1B params

Tensor type

F32

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Mothersuperior/YuE2-Vae-merge-0.666

Base model

m-a-p/YuE2-Vae

Finetuned

(1)

this model