BF16/FP16 Version
#3
by UbuUbu - opened
Hello, thanks for you awesome work, could you release a BF16/FP16 version so I can make a MXFP8 quant.
MXFP8 deliver faster inference on 50xx gpu with almost FP16 quality and FP8 size, it's almost magic :D
Hello, thanks for you awesome work, could you release a BF16/FP16 version so I can make a MXFP8 quant.
MXFP8 deliver faster inference on 50xx gpu with almost FP16 quality and FP8 size, it's almost magic :D
Here you can get it now 🤝✌️: https://huggingface.co/wikeeyang/Magic-Wan-T2IV-V3/resolve/main/Magic-Wan-Image-V3-mxfp8mixed.safetensors
Yeaahhh downloading right now ♥
A BF16 version would still be really helpful for more exotic quantizations (INT8), and to finetune on.