[thepasswordis]

Is it possible to run this on something other than google's cloud platform?

[Ari_Rahikkala]

I'm running it comfortably on my 3090, although it's a really snug fit for the VRAM, and that's with a number of fixes to significantly reduce its memory use from https://github.com/AeroScripts/mesh-transformer-jax .

[homedepotdave]

Out of curiosity, how fast are your inferences with this setup?

[Ari_Rahikkala]

With the defaults of per_replica_batch=1, seq=2048 and gen_len=512, a completion takes about 20 seconds.

I'm not sure yet what settings I'll end up with if I decide to play with this more. per_replica_batch=3, seq=1024, gen_len=64 would give an experience roughly similar to the AI Dungeon that I'm used to, though less clever than the Dragon model, and a bit slower at about 10 seconds per batch.

[londons_explore]

It's design is pretty specific to Google's TPU.

If you ran it elsewhere, you'd likely need to tweak the design or suffer quite a big performance penalty.

[devetec]

Nope, probably not. It's using JAX, which works on both GPUs and TPUs.

[londons_explore]

All the constants in the design will be tuned to fit perfectly in TPU hardware dimensions.