Sayak Paul
@RisingSayak
ML at Hugging Face đ¤
Joined May 2012
125 Following 24.8K Followers
Tensor parallel loading in đ¤ Diffusers got a massive upgrade đĨ
On Flux.2-Dev DiT, with a TP degree of 4 (A10G):
âĸ 30.4s â 12.5s loading
âĸ 64.1 â 6.8 GB peak CPU memory/rank
~2.4Ã faster loading. ~89% less CPU memory.
Check it out:
Show more
