Highly experimental prune of GLM 4.6V from 128 experts to 96. Vision works, as tested in GGUF format. I would greatly appreciate feedback.

Safetensors

Model size

83B params

Tensor type

BF16

F32

Model tree for blascotobasco/GLM-4.6V-96E

Base model

zai-org/GLM-4.6V

Finetuned

(6)

this model

Quantizations