tshrjn / finetune-qa Goto Github PK

Came here from https://github.com/PyTorchLightning/pytorch-lightning, the code won't run as it hardcodes creator's wandb username/entity:

python main.py --gpus 1 --qa_model distilroberta-base --workers 20 --bs 5 --max_epochs 10
[...]
<transformers.tokenization_roberta.RobertaTokenizerFast object at 0x7f3ad04f5ad0>
wandb: Tracking run with wandb version 0.9.2
wandb: Run data is saved locally in wandb/run-20200706_233437-3sbo0kxn
wandb: ERROR Error while calling W&B API: permission denied (<Response [403]>)
wandb: ERROR Launch exception: Permission denied to access nlp/qa/3sbo0kxn
wandb: ERROR To disable wandb syncing set WANDB_MODE=dryrun
wandb: ERROR W&B process (PID 10015) did not respond
wandb: ERROR W&B process failed to launch, see: wandb/debug.log

one needs to edit main.pyand replace: entity='nlp' in:

wandb_logger = WandbLogger(project='qa', entity='nlp', ...)

with entity='YOUR WANDB USERNAME'.

And, of course, if you haven't used wandb yet, you need to create an account first and get the API key.

I, of course, realize that perhaps this is just an example code, so it was never supposed to be runable by others as is... in which case please ignore this issue.

Please add the code that generates the predictions and n_best_predictions file.

Hey,
I cloned the repo, pip installed the requirements, ran main, and got:
TypeError: special token bos_token has to be either str or AddedTokenFast but got: <class 'dict'>

I needed to upgrade the transformers on my own and it got solved.

hey, it seems like the script doesn't split the input examples into segments and doesn't do evaluation. Did I miss anything?