Problems with inference

#38
by Kirill200223 - opened

I'm trying to load the model in the simplest way through torch, but I'm always stuck on
Loading weights: 778it [04:05, 3.17it/s, Materializing param=lm_head.weight]
the graphics cards are loaded, but the process hangs endlessly in this place. Moreover, the same code works without errors in 1 out of every 10 cases. Setup: 4 A100 80GB

I've tried running it on different servers in different environments, and the problem is the same everywhere. It's like accelerate freezes in deadlock.

Sign up or log in to comment