Any chance for an fp16 .safetensors version?
#1
by CCP6 - opened
Hello,
Thank you for your efforts on making these "heavy" new SOTA model faster to generate. Unfortunately the Mochi Fast model you share here now is 40gb putting it out of VRAM range of any consumer GFX card. Is there any possibility you could convert/host an fp16 version? Thanks for reading/considering!!