Unable to find .ckpt. file

[nishithbenhur]

This is a repository of pretrained Japanese BERT models. The pretrained models are available along with the source code of pretraining.

Hi Team,
As mentioned above, i am unable to find the .ckpt file . My intention is is to host the model as a service and i need the below files for the same . Can you let me know on this.

├── model.ckpt.data-00000-of-00001
├── model.ckpt.index
├── model.ckpt.meta

[nishithbenhur]

If these are the models included as below then which one can be user for FAQ answering. Assume i have a dataset of 100 Questions and corresponding answers. This is for a Chatbot application

BERT-base_mecab-ipadic-bpe-32k.tar.xz (2.1GB)
BERT-base_mecab-ipadic-bpe-32k_whole-word-mask.tar.xz (2.1GB)
BERT-base_mecab-ipadic-char-4k.tar.xz (1.6GB)
BERT-base_mecab-ipadic-char-4k_whole-word-mask.tar.xz (1.6GB)

[singletongue]

The model files are included in tar.xz archive files.
You can try any of them, but I'm afraid that 100 QA pairs might be too less for fine-tuning a model.

[nishithbenhur]

If these are the models included as below then which one can be user for FAQ answering. Assume i have a dataset of 100 Questions and corresponding answers. This is for a Chatbot application

BERT-base_mecab-ipadic-bpe-32k.tar.xz (2.1GB)
BERT-base_mecab-ipadic-bpe-32k_whole-word-mask.tar.xz (2.1GB)
BERT-base_mecab-ipadic-char-4k.tar.xz (1.6GB)
BERT-base_mecab-ipadic-char-4k_whole-word-mask.tar.xz (1.6GB)

Dear Suzuki,

Thank you for the reply.
I thought the model is readily available for using . Which model can be readily used for FAQ dataset of 300 Questions

[singletongue]

There are no difference between the models in usability.
If you're still unsure which model to use, why not give it a try with the first one?

For a quick trial, I recommend to use our models in Hugging Face's Transformers package.
Simple demo: https://twitter.com/huggingface/status/1205283603128758277

[nishithbenhur]

There are no difference between the models in usability.
If you're still unsure which model to use, why not give it a try with the first one?

For a quick trial, I recommend to use our models in Hugging Face's Transformers package.
Simple demo: https://twitter.com/huggingface/status/1205283603128758277

Thank you. I will try the first one and give my feedback and observations so that rest of the people are benefited too.

1. Also could you tell me the differences between as i need to know which model can be used for FAQ answering in Chatbot

• BERT-base_mecab-ipadic-bpe-32k.tar.xz (2.1GB)
• BERT-base_mecab-ipadic-bpe-32k_whole-word-mask.tar.xz (2.1GB)
• BERT-base_mecab-ipadic-char-4k.tar.xz (1.6GB)
• BERT-base_mecab-ipadic-char-4k_whole-word-mask.tar.xz (1.6GB)

2. How can i get the .ckpt file for bert-base-japanese so that i can host it a HTTP API service?

[xeisberg]

@nishithbenhur

1. I am not sure. Character may focus on each individual character in opposition to words(tokenised) in Japanese which are often made up of several characters. I do not know about difference between whole word mask and not whole word mask, it would be interesting to know about the differences in performance.

2. You can find the files here: https://www.nlp.ecei.tohoku.ac.jp/%7Em-suzuki/bert-japanese/