Dockerで日本語BERTを使ってみる(旧版)

$ docker pull ishizakiyuko/japanese_bert_trial:1.0.1
1.0.1: Pulling from ishizakiyuko/japanese_bert_trial
Digest: sha256:7d3aaf385e7c2ecc810eaf96098bb522b945fcdeb70e104e0a5cc3e3a8fd7182
Status: Downloaded newer image for ishizakiyuko/japanese_bert_trial:1.0.1
docker.io/ishizakiyuko/japanese_bert_trial:1.0.1

$ docker image ls
REPOSITORY                         TAG       IMAGE ID       CREATED       SIZE
ishizakiyuko/japanese_bert_trial   1.0.1     0d5c7dc32e9d   3 years ago   10.5GB

$ docker run -p 8888:8888 -d ishizakiyuko/japanese_bert_trial:1.0.1

$ docker container ls
CONTAINER ID   IMAGE                                    COMMAND                  CREATED          STATUS          PORTS                                       NAMES
e0ba28ed92ee   ishizakiyuko/japanese_bert_trial:1.0.1   "/bin/sh -c '/opt/co…"   50 seconds ago   Up 50 seconds   0.0.0.0:8888->8888/tcp, :::8888->8888/tcp   musing_jang

$ docker logs e0ba28ed92ee
[I 02:23:09.614 NotebookApp] Writing notebook server cookie secret to /home/jovyan/.local/share/jupyter/runtime/notebook_cookie_secret
[I 02:23:09.730 NotebookApp] Serving notebooks from local directory: /home/jovyan
[I 02:23:09.730 NotebookApp] The Jupyter Notebook is running at:
[I 02:23:09.730 NotebookApp] http://e0ba28ed92ee:8888/?token=c909ce4c1770455b12e6ef46d4b13ef7733190affe22b510
[I 02:23:09.730 NotebookApp]  or http://127.0.0.1:8888/?token=c909ce4c1770455b12e6ef46d4b13ef7733190affe22b510
[I 02:23:09.730 NotebookApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
[C 02:23:09.733 NotebookApp] 

    To access the notebook, open this file in a browser:
        file:///home/jovyan/.local/share/jupyter/runtime/nbserver-7-open.html
    Or copy and paste one of these URLs:
        http://e0ba28ed92ee:8888/?token=c909ce4c1770455b12e6ef46d4b13ef7733190affe22b510
     or http://127.0.0.1:8888/?token=c909ce4c1770455b12e6ef46d4b13ef7733190affe22b510
[I 02:23:26.334 NotebookApp] 302 GET / (172.17.0.1) 0.48ms
[I 02:23:26.338 NotebookApp] 302 GET /tree? (172.17.0.1) 0.78ms

$ google-chrome &

import os
import torch
from transformers import BertForMaskedLM, BertConfig, BertTokenizer
from pyknp import Juman

BASE_PATH = './Japanese_L-12_H-768_A-12_E-30_BPE_WWM_transformers'
BERT_CONFIG = 'config.json'
BERT_MODEL = 'pytorch_model.bin'
VOCAVULARY_LIST = 'vocab.txt'

jumanpp = Juman()

# 形態素解析
text = 'どんなに勉強しても全然頭が良くならない'
result = jumanpp.analysis(text)
tokenized_text =[mrph.midasi for mrph in result.mrph_list()]
print(tokenized_text)

# Mask 
tokenized_text.insert(0, '[CLS]')
tokenized_text.append('[SEP]')

masked_index = 6 # Maskしたいtextのindex 
tokenized_text[masked_index] = '[MASK]'
print(tokenized_text)

# Bert model
config = BertConfig.from_json_file(os.path.join(BASE_PATH, BERT_CONFIG))
model = BertForMaskedLM.from_pretrained(os.path.join(BASE_PATH, BERT_MODEL), config=config)
tokenizer = BertTokenizer(os.path.join(BASE_PATH, VOCAVULARY_LIST), do_lower_case=False, do_basic_tokenize=False)

# token化
indexed_tokens = tokenizer.convert_tokens_to_ids(tokenized_text)
tokens_tensor = torch.tensor([indexed_tokens])
print(tokens_tensor)

# 予測
model.eval()

tokens_tensor = tokens_tensor.to('cpu')
model.to('cpu')

with torch.no_grad():
    outputs = model(tokens_tensor)
    predictions = outputs[0]

_, predicted_indexes = torch.topk(predictions[0, masked_index], k=5)
predicted_tokens = tokenizer.convert_ids_to_tokens(predicted_indexes.tolist())
print(predicted_tokens)