All I found was this https://twitter.com/kliu128/status/1623472922374574080?s=21

This describes a LLM + reinforcement learning hybrid that has been trained to navigate webpages for arbitrary tasks. I’m not sure how far away this is, or if it already exists. Someone below mentioned an action transformer which may be related

I don’t know much about this either but it seems you’re right since there’s not a one size fits all solution for anything. But at the same time, it’s not very helpful advice

The ecosystems are very dissimilar so it’s not easy to deploy a model in both

I don’t know much about this topic but to me, it seems there is a difference between a SOTA model developed for a paper and a model ready for deployment. Seems torch is easier to dev while tf uses lazy evaluation and is therefore lighter. Could be wrong

Yeah I mentioned that one in my description. Cool paper. Not as good as FNO, though, even if they report otherwise

This only includes FNO, Unet, and PINN

I want to compare FNO to other SOTA that have been published since ICLR 21 on the 2d NS task from the FNO paper - that is, predict the scalar vorticity of NS on a regular spatial temporal grid for 20-40 time steps into the future given the first 10 solutions

I’m looking for competitors for FNO - any recommendations on models that would be a fair comparison for solving 2d time evolving NS?

I’m looking for competitors for FNO - any recommendations for models that would be a fair comparison?