batterydata's picture
Update README.md
532c3f8
metadata
language: en
tags: question answering
license: apache-2.0
datasets:
  - squad
  - batterydata/battery-device-data-qa
metrics: squad

BatterySciBERT-cased for QA

Language model: batteryscibert-cased Language: English
Downstream-task: Extractive QA
Training data: SQuAD v1 Eval data: SQuAD v1 Code: See example Infrastructure: 8x DGX A100

Hyperparameters

batch_size = 32
n_epochs = 3
base_LM_model = "batteryscibert-cased"
max_seq_len = 386
learning_rate = 2e-5
doc_stride=128
max_query_length=64

Performance

Evaluated on the SQuAD v1.0 dev set.

"exact": 79.66,
"f1": 87.43,

Evaluated on the battery device dataset.

"precision": 65.09,
"recall": 84.56,

Usage

In Transformers

from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline

model_name = "batterydata/batteryscibert-cased-squad-v1"
# a) Get predictions
nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)
QA_input = {
    'question': 'What is the electrolyte?',
    'context': 'The typical non-aqueous electrolyte for commercial Li-ion cells is a solution of LiPF6 in linear and cyclic carbonates.'
}
res = nlp(QA_input)
# b) Load model & tokenizer
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

Authors

Shu Huang: sh2009 [at] cam.ac.uk

Jacqueline Cole: jmc61 [at] cam.ac.uk

Citation

BatteryBERT: A Pre-trained Language Model for Battery Database Enhancement