Are these possible to quantize? (lynx models)
#21
by
patientxtr
- opened
https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lynx
these models (at least one ip_layers and one full_layers needs to be used with main wan model and at the normal size without using gguf (so all fp8 or fp16) they are too big for 16gb vram. We already have gguf's for wan so would it be possible to get gguf's for the lynx models so we can use them in wrapper in comfyui ?
is it a lora or adapter? 14b f16 should not be 4GB that small
did them anyway thanks to a bit of modification on your script, for anyone insterested : https://huggingface.co/patientxtr/Wan2_1-T2V-14B-Lynx-GGUF