| Property | Value |
|---|---|
| Parameter Count | 111M |
| License | CC-BY-SA-4.0 |
| Author | sonoisa |
| Base Model | cl-tohoku/bert-base-japanese-whole-word-masking |
What is sentence-bert-base-ja-mean-tokens-v2?
This is an improved Japanese Sentence-BERT model that leverages the MultipleNegativesRankingLoss function for better sentence embedding generation. Built upon the successful cl-tohoku/bert-base-japanese-whole-word-masking architecture, this v2 model demonstrates 1.5-2 points higher accuracy compared to its predecessor on private datasets.
Implementation Details
The model utilizes PyTorch framework and requires fugashi and ipadic for inference. It implements mean pooling strategy for generating sentence embeddings and supports batch processing for efficient computation.
- Improved loss function using MultipleNegativesRankingLoss
- Built on BERT-base Japanese whole word masking
- Supports batch processing with customizable batch sizes
- Implements efficient mean pooling strategy
Core Capabilities
- Japanese sentence embedding generation
- Semantic similarity comparison
- Feature extraction for Japanese text
- Efficient batch processing of sentences
Frequently Asked Questions
Q: What makes this model unique?
This model stands out due to its improved training approach using MultipleNegativesRankingLoss and its specific optimization for Japanese language processing, showing measurable improvements over the v1 model.
Q: What are the recommended use cases?
The model is ideal for Japanese text processing tasks including semantic similarity comparison, document classification, and feature extraction for downstream NLP tasks.