The question-answering-tutorial's discuss from vgaraujov

vgaraujov / question-answering-tutorial Goto Github PK

View Code? Open in Web Editor NEW

Tutorial of Question Answering using SQuAD in English and Spanish with BERT and BiDAF.

Jupyter Notebook 100.00%

question-answering-tutorial's Issues

Error when try this

Gives an error when running this format with transformers 3.5.1-
as huggingface -transformers update their script i found it in their legacy folder then download in colab & run

!export SQUAD_DIR=/content/dataset \
&& python /content/run_squad.py \
  --model_type bert \
  --model_name_or_path sagorsarker/bangla-bert-base \
  --do_train \
  --do_eval \
  --do_lower_case \
  --train_file $SQUAD_DIR/bengali.json \
  --predict_file $SQUAD_DIR/val_data.json \
  --per_gpu_train_batch_size 6 \
  --learning_rate 3e-5 \
  --num_train_epochs 5.0 \
  --max_seq_length 384 \
  --doc_stride 128 \
  --output_dir /content/model_output \
  --save_steps 5000 \
  --threads 4 \

*error

Evaluating:   0% 0/16 [00:00<?, ?it/s]
Traceback (most recent call last):
  File "/content/run_squad.py", line 830, in <module>
    main()
  File "/content/run_squad.py", line 819, in main
    result = evaluate(args, model, tokenizer, prefix=global_step)
  File "/content/run_squad.py", line 324, in evaluate
    output = [to_list(output[i]) for output in outputs.to_tuple()]
AttributeError: 'tuple' object has no attribute 'to_tuple'

I've specific questions. There is a model sagorsarker/bangla-bert-base already available for Bengali language, also a squad like dataset from google tydiqa contains questions of Bengali. So, run_squad.py is just simply work for me ? as simple as that or i Use run_squad_trainer.py ?(updated script) from huggingface repo.

Thank you.

Recommend Projects