EXL3 6.66bpw H8 quant, tested using it with 24k q8 context within 24GB VRAM.
Recipe based on
knifeayumu/Cydonia-v1.2-Magnum-v4-22B. Just an update to those who are interested.
Wan-AI/Wan2.2-I2V-A14B was used to turn this
image from the previous merge (slightly cropped) to an animation utilising
lightx2v/Wan2.2-Lightning for faster generation and
pollockjj/ComfyUI-MultiGPU nodes.
This is a merge of pre-trained language models created using
mergekit.
This model was merged using the
SLERP merge method.
1models:
2 - model: TheDrummer/Cydonia-24B-v4.1
3 - model: Doctor-Shotgun/MS3.2-24B-Magnum-Diamond
4merge_method: slerp
5base_model: TheDrummer/Cydonia-24B-v4.1
6parameters:
7 t: [0.1, 0.3, 0.6, 0.3, 0.1]
8dtype: bfloat16
