Model Overview

This is an uncased question answering model with a BERT Base encoder finetuned on dataset SQuADv1.1 [1]. With Question Answering, or Reading Comprehension, given a question and a passage of content (context) that may contain an answer for the question, the model predicts the span within the text with a start and end position indicating the answer to the question.

Model Architecture

The current version of the question answering model The model is based on the architecture presented in "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding" paper [2]. In this particular instance, the model has 24 Transformer blocks. On top of that it is using a span prediction head, that is equivalent to token classification with 2 classes: one for the start of the span and one for the end of the span. All model parameters are jointly fine-tuned on the downstream task. More specifically, an input text is fed to the BERT encoder model, and the output states are further fed to the span prediction.

Training

The model was trained with NeMo BERT Base uncased checkpoint.

Dataset

The model was trained on SQuADv1.1 [1] corpus for question answering. For datasets like SQuAD 1.1, this model only supports cases where the answer is contained in the context.

Performance

Evaluation on the SQuAD1.1 dev set:

Exact Match 82.78%

F1 89.97%

How to use this model

The model is available for use in the NeMo toolkit [3], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.

Automatically load the model from NGC

import nemo
import nemo.collections.nlp as nemo_nlp
model = nemo_nlp.models.question_answering.qa_model.QAModel.from_pretrained(model_name="qa_squadv1.1_bertbase")

Inference

python [NEMO_GIT_FOLDER]/examples/nlp/question_answering/question_answering_squad.py do_training=false pretrained_model=qa_squadv1.1_bertbase model.validation_ds.file=[SOURCE_FILE] model.dataset.do_lower_case=true 

Input

The model takes a Json file as input that follows the SQuAD format.

Output

The model outputs a JSON file as output for prediction and n-Best list.

Limitations

The length of the input text is currently constrained by the maximum sequence length of the uncased encoder model, which is 512 tokens after tokenization.

References

[1] https://rajpurkar.github.io/SQuAD-explorer/

[2] https://arxiv.org/pdf/1810.04805.pdf

[3] NVIDIA NeMo Toolkit

Licence

License to use this model is covered by the NGC TERMS OF USE unless another License/Terms Of Use/EULA is clearly specified. By downloading the public and release version of the model, you accept the terms and conditions of the NGC TERMS OF USE.