Cristian García Romero

I've been dealing with the same problem, but for other language pair (et-en). I solved my problem adding the following flags to `fairseq-train`: ```bash --lr-scheduler inverse_sqrt --warmup-updates 8000 --warmup-init-lr 1e-7...

Hi, @TheodorePeng, Yes, the BLEU value was close to 0 in training and evaluation. The problem was related to the fact that I was not using a LR scheduler. As...

I've run into this problem as well. In case it helps someones, you can modify this timeout from the installation: ```bash f=$(python -c 'import service_streamer as ss; print(ss.service_streamer.__file__)') echo "$f"...

Hi! I've running into this problem. I've observed that decreasing `--mini-batch` kind of mitigates the problem. But why does this problem happens? Why does the memory usage stop increasing? Does...

Same issue here. Any solution?

Pity :/ It happens intermittently to me. You can see in [1](https://github.com/bitextor/bitextor/runs/7595849987?check_suite_focus=true) and [2](https://github.com/bitextor/bitextor/runs/7611585960?check_suite_focus=true) that the job "Tests combining dictionaries and MT" is cancelled in one and not in the...

Thank you! And I forgot to mention that I used CPU in order to train the models, not GPU.

I ran the same tests multiple times and now I'm getting the same models! Thank you!

By the way, I'm not sure about Bicleaner AI regarding the reproducible training (I'll check it out, eventually), but I found this in the code (https://github.com/bitextor/bicleaner-ai/blob/39326eab2026b168fdb487cdc02ffe873f3345f2/bicleaner_ai/bicleaner_ai_train.py#L115): ```python os.environ["PYTHONHASHSEED"] = str(args.seed)...

I agree that I'd be useful to have a reference somewhere in the github page. The paper you mentioned is a little bit old (I haven't read it, but I'm...