roberta-base-squad2 for Extractive QA on COVID-19

Overview

Language model: deepset/roberta-base-squad2
Language: English
Downstream-task: Extractive QA
Training data: SQuAD-style CORD-19 annotations from 23rd April
Code: See an example extractive QA pipeline built with Haystack
Infrastructure: Tesla v100

Hyperparameters

batch_size = 24
n_epochs = 3
base_LM_model = "deepset/roberta-base-squad2"
max_seq_len = 384
learning_rate = 3e-5
lr_schedule = LinearWarmup
warmup_proportion = 0.1
doc_stride = 128
xval_folds = 5
dev_split = 0
no_ans_boost = -100

license: cc-by-4.0

Performance

5-fold cross-validation on the data set led to the following results:

Single EM-Scores: [0.222, 0.123, 0.234, 0.159, 0.158]
Single F1-Scores: [0.476, 0.493, 0.599, 0.461, 0.465]
Single top\_3\_recall Scores: [0.827, 0.776, 0.860, 0.771, 0.777]
XVAL EM: 0.17890995260663506
XVAL f1: 0.49925444207319924
XVAL top\_3\_recall: 0.8021327014218009

This model is the model obtained from the third fold of the cross-validation.

Usage

In Haystack

Haystack is an AI orchestration framework to build customizable, production-ready LLM applications. You can use this model in Haystack to do extractive question answering on documents. To load and run the model with Haystack:

# After running pip install haystack-ai "transformers[torch,sentencepiece]"

from haystack import Document
from haystack.components.readers import ExtractiveReader

docs = [
    Document(content="Python is a popular programming language"),
    Document(content="python ist eine beliebte Programmiersprache"),
]

reader = ExtractiveReader(model="deepset/roberta-base-squad2")
reader.warm_up()

question = "What is a popular programming language?"
result = reader.run(query=question, documents=docs)
# {'answers': [ExtractedAnswer(query='What is a popular programming language?', score=0.5740374326705933, data='python', document=Document(id=..., content: '...'), context=None, document_offset=ExtractedAnswer.Span(start=0, end=6),...)]}

For a complete example with an extractive question answering pipeline that scales over many documents, check out the corresponding Haystack tutorial.

In Transformers

from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline

model_name = "deepset/roberta-base-squad2"

# a) Get predictions
nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)
QA_input = {
    'question': 'Why is model conversion important?',
    'context': 'The option to convert models between FARM and transformers gives freedom to the user and let people easily switch between frameworks.'
}
res = nlp(QA_input)

# b) Load model & tokenizer
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)