Hi,
I noticed that there's no Universal_V3 model. Is this because V3 is too small to do well in the generalized case? Or is it possible it does quite well but simply hasn't been tried / trained yet?
Was Universal_V3 likewise trained for 2,500K steps? How long did this take for such a large dataset and was it also done with 2 V100 GPUs?
Couldn't find the info on the paper, so hope @jik876 can provide the answers!
Hi, I noticed that there's no Universal_V3 model. Is this because V3 is too small to do well in the generalized case? Or is it possible it does quite well but simply hasn't been tried / trained yet?
Was Universal_V3 likewise trained for 2,500K steps? How long did this take for such a large dataset and was it also done with 2 V100 GPUs?
Couldn't find the info on the paper, so hope @jik876 can provide the answers!