README.md

metadata

license: cc-by-sa-3.0
language: ja
tags:
  - question-answering
  - extractive-qa
pipeline_tag:
  - None
datasets:
  - SkelterLabsInc/JaQuAD
metrics:
  - Exact match
  - F1 score

BERT base Japanese - JaQuAD

Description

A Japanese Question Answering model fine-tuned on JaQuAD. Please refer BERT base Japanese for details about the pre-training model. The codes for the fine-tuning are available at SkelterLabsInc/JaQuAD

Evaluation results

On the development set.

{"f1": 77.35, "exact_match": 61.01}

On the test set.

{"f1": 78.92, "exact_match": 63.38}

Usage

from transformers import AutoModelForQuestionAnswering, AutoTokenizer

question = 'アレクサンダー・グラハム・ベルは、どこで生まれたの?'
context = 'アレクサンダー・グラハム・ベルは、スコットランド生まれの科学者、発明家、工学者である。世界初の>実用的電話の発明で知られている。'

model = AutoModelForQuestionAnswering.from_pretrained(
    'SkelterLabsInc/bert-base-japanese-jaquad')
tokenizer = AutoTokenizer.from_pretrained(
    'SkelterLabsInc/bert-base-japanese-jaquad')

inputs = tokenizer(
    question, context, add_special_tokens=True, return_tensors="pt")
input_ids = inputs["input_ids"].tolist()[0]
outputs = model(**inputs)
answer_start_scores = outputs.start_logits
answer_end_scores = outputs.end_logits

# Get the most likely beginning of answer with the argmax of the score.
answer_start = torch.argmax(answer_start_scores)
# Get the most likely end of answer with the argmax of the score.
# 1 is added to `answer_end` because the index pointed by score is inclusive.
answer_end = torch.argmax(answer_end_scores) + 1

answer = tokenizer.convert_tokens_to_string(
    tokenizer.convert_ids_to_tokens(input_ids[answer_start:answer_end]))
# answer = 'スコットランド'

License

The fine-tuned model is licensed under the CC BY-SA 3.0 license.

Citation

TBA