Code of pre-train about genre HOT 3 CLOSED

@Valdegg mGENRE is a simple sequence to sequence model so if you want to fine-tune on a specific language you would just need to use fairseq-train specifying the model checkpoint and the data folder. See this for an example https://github.com/facebookresearch/GENRE/blob/main/scripts_genre/train.sh

from genre.

I second this - would be great to have access to the codes, to build upon this nice project.

I want to fine-tune mGENRE to train it on a specific language and I'm worried that the codes I make myself won't do the job.

from genre.

Hello, you are right I left the description of these files but they are not provided. Unfortunately, I do not have those files anymore (besides, run_bart_slurm.py includes internal code we cannot release---also it is useless as it has specific commands to run training on a FAIR distributed cluster).

run_training.sh simply launches fairseq-train with the (m)BART pre-trained model and specifying the data folder.

from genre.