| Property | Value |
|---|---|
| License | Apache 2.0 |
| Architecture | BERT Base (12 layers, 768 hidden, 12 heads) |
| Training Data | CC-100 (74.3GB) + Wikipedia (4.9GB) |
| Vocabulary Size | 32,768 tokens |
What is bert-base-japanese-v3?
bert-base-japanese-v3 is a specialized Japanese language model based on the BERT architecture, developed by Tohoku NLP. It implements word-level tokenization using the Unidic 2.1.2 dictionary, combined with WordPiece subword tokenization, making it particularly effective for Japanese text processing.
Implementation Details
The model underwent a sophisticated training process, with 1M steps on CC-100 corpus followed by 1M steps on Wikipedia data. It utilizes Cloud TPUs (v3-8) for training and implements whole word masking for the masked language modeling objective.
- Word-level tokenization using MeCab with Unidic 2.1.2 dictionary
- Subword tokenization using WordPiece algorithm
- Training corpus combining CC-100 (392M sentences) and Wikipedia (34M sentences)
- Whole word masking implementation for better contextual understanding
Core Capabilities
- Advanced Japanese text processing and understanding
- Efficient tokenization handling both word and subword levels
- Robust performance on masked language modeling tasks
- Suitable for various Japanese NLP applications
Frequently Asked Questions
Q: What makes this model unique?
This model's uniqueness lies in its combination of word-level tokenization using Unidic 2.1.2 and whole word masking, along with its extensive training on both CC-100 and Wikipedia data, making it particularly effective for Japanese language tasks.
Q: What are the recommended use cases?
The model is well-suited for Japanese natural language processing tasks, including text classification, named entity recognition, and masked language modeling. It's particularly effective for applications requiring deep understanding of Japanese text structure.