BERT for QA fine-tuned on SQuAD v1.1

Release with a version of BERT model trained on SQuAD 1.1 following last updates on cdQA modules.

This model is CPU/GPU agnostic. If the model is loaded in a machine with support for CUDA it will automatically send the model to GPU for computations.

This version of the model achieves 81.3% EM and 88.7% F1-score on SQuAD 1.1 dev.

Distilbert for QA fine-tuned on SQuAD v1.1 with Knowledge Distillation

Release with a version of DistilBERT model trained on SQuAD 1.1 using Knowledge Distillation and bert-large-uncased-whole-word-masking-finetuned-squad as a teacher.

This version of Distilbert achieves 80.1% EM and 87.5% F1-score (vs. 81.2% EM and 88.6% F1-score for our version of BERT), while being much faster and lighter.

Version available only with sklearn wrapper.

GPU version of BERT QA fine-tuned on SQuAD v1.1 (with sklearn wrapper)

Release with a version of BERT model trained on SQuAD 1.1 runnable on GPU.

Version available only with sklearn wrapper (bert_qa_vGPU-sklearn.joblib)

CPU version of BERT QA fine-tuned on SQuAD v1.1 (with sklearn wrapper)

Release with a version of BERT model trained on SQuAD 1.1 runnable on CPU.

Version available in two formats:

Pytorch model: bert_qa_vCPU.bin and config.json

Sklearn wrapper: bert_qa_vCPU-sklearn.joblib

BNP Paribas Newsroom dataset v.1.1

A dataset of 3675 BNP Paribas public articles with content and metadata available on the "Newsroom" page of the official BNP Paribas website.

• bnpp_newsroom-v1.1.csv: contains all articles paragraphs, non-filtered.