firqaaa

/

indo-dpr-ctx_encoder-single-squad-base

like 0

Hello Mas Firqa,

I hope you are doing well. I have been exploring the fine-tuned DPR models and would like to ask a few questions regarding the training process and dataset preparation:

1. How was the process of the dataset transformed from the SQuAD 2.0 format into the DPR training format?
2. How many training instances were used after the dataset preparation process for fine-tuning the DPR model?
3. Was the model fine-tuned using IndoBERT? If so, which IndoBERT paper or pretrained model did you refer to?
4. Do the context encoder and question encoder use the same encoder architecture or weights?

Thank you for your time and for sharing your work. I really appreciate it and will look forward to your response.