Roman Grundkiewicz

With `--mini-batch-fit` these two parameters still determine how many sentences is loaded for preparing batches, it's `mini-batch` x `maxi-batch`.

Please check if you don't have very long lines in your input and clean them, or check a small mini-batch size. You could also set smaller `max-length` and `max-length-crop=true`, but...

Ping @kpu

The options "factors-combine" and "factors-dim-emb" have been added recently and I think it seems the check whether they are provided or not is missing. The issue wasn't caught because most...

By "GNU version is 7.5", do you mean GCC/G++ 7.5? What version of cmake do you use? (the output of `cmake --version`)

Marian compiles with CUDA 10.2 (driver version 440.33.01) and GCC 5.4.0 without issues in my case (Ubuntu 16.04). So I guess that GCC 7.5 should also work with CUDA 10.2...

Hi Toms, I've never used it and don't know what it actually does. Pinging @emjotde, who may know. It might be that it was an experimental feature.

An alternative to enabling `--max-length-crop` by default might be generating an empty translation if the decoder sees an input sentence exceeding `--max-length`, and I thought that it already does that....

I don't know if this is correct or optimal. I will need to take a closer look. I remember that I have updated the adaptive branch and haven't finished, but...

I updated the description to make the proposed approach clearer.