node-red-contrib-bert-tokenizer

Implement BERT tokenizer as a custom Node-RED custom node

Installation

npm install node-red-contrib-bert-tokenizer

Usage

The input payload is text (i.e. words, sentences, etc) to be converted into input feature for the BERT model. The input is tokenized and no longer than 126 tokens. Each name array in the output object has 128 numbers. See BERT-Tokenizer for an example output.

Optional node property:

URL: vocabulary url must start with:

1. leave empty to load the default
2. http(s):// to fetch the model from web.
3. file:// to load the vocaburlary file on the server.