Timothee Mickus
Timothee Mickus
- [ ] break down accuracy and validation loss per task when reporting stats - [ ] introduce validation-based LR scheduling ([reduce LR on plateau](https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.ReduceLROnPlateau.html)) that's properly adjusted depending on...
- [ ] Redesign parameter initialization so as to ensure a consistent and systematic design - [ ] Allow non-random initialization: - [ ] from huggingface - [ ] from...
In the long run, rather than having our custom transforms for data cleaning (as suggested b y #13), it would be better to leave it to a relevant third party,...
Setting a shard_size different from 0 currently causes the model to duplicate the test set. The simplest way to circumvent this at the moment is to set the flag -shard_size...
General PR for the HF/mammoth integration project, will also put some doc/pointers here
Description of the problem --------------------------------------- Attested on a live competition on the production LISP UP Saclay server. One of our participants is not able to submit through the platform: every...
Currently logs are overwhelming and not human-readable. Would be great to sift through the current messages and set appropriate logging levels (also remove sneaky prints that are surely still around).
Given the scope of experiments, it should be enough to have a global flag in the config, and then after the encoder is created here: https://github.com/Helsinki-NLP/mammoth/blob/1e6b03a6ce3853131d093d73bf91fdad1ed2a48d/mammoth/model_builder.py#L361 iterate over its parameters...