roberta-squad
roberta-squad copied to clipboard
error trying to train squad /tokenizer/roberta.so: undefined symbol: _Py_ZeroStruct
Thank you for sharing the code for fine-tuning Roberta.
I've been trying to train Roberta-base on squad v2 and I am getting the following error with this command:
python ./fairseq_train.py ./squad/v2 --restore-file /home./roberta.base/model.pt --reset-optimizer --reset-dataloader --reset-meters --no-epoch-checkpoints --no-last-checkpoints --no-save-optimizer-state --task squad2 --max-positions 512 --arch roberta_qa --dropout 0.1 --attention-dropout 0.1 --weight-decay 0.01 --criterion squad2 --optimizer adam --adam-betas '(0.9, 0.98)' --adam-eps 1e-06 --clip-norm 0.0 --lr-scheduler polynomial_decay --lr 1.5e-5 --memory-efficient-fp16 --warmup-updates 326 --total-num-update 5430 --max-sentences 3 --required-batch-size-multiple 1 --update-freq 2 --max-update 5430 --lr_decay 1.0 --ddp-backend=no_c10d --num-workers=32
from tokenizer.roberta import RobertaTokenizer, MASKED, NOT_MASKED, IS_MAX_CONTEXT, NOT_IS_MAX_CONTEXT
ImportError: /home/roberta-squad/tokenizer/roberta.so: undefined symbol: _Py_ZeroStruct
(
Would appreciate any help on how to resolve.
Thanks
@shairoz-deci, did you have any luck using this code?