This is a copy of KaraKaraWitch/CavesOfQwen3 with the number of active experts turned up to 16.

This will actually make performance worse if you just use it out of the box (don't let viral reddit posts fool you, turning up experts is one of the first things a lot of us tried to try to stabilize this model). The goal is to try training with more experts active so the model learns to use them properly, hence prepping a version with that already in the config for easy loading.

Downloads last month
59
Safetensors
Model size
30.5B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for allura-forge/coq3-16e-mixed-data-merged