ko-sbert-nli
| Property | Value |
|---|---|
| Author | jhgan |
| Downloads | 28,533 |
| Paper | Research Paper |
| Embedding Dimension | 768 |
| Performance | 82.24% Cosine Pearson on KorSTS |
What is ko-sbert-nli?
ko-sbert-nli is a specialized Korean language model based on SBERT (Sentence-BERT) architecture, designed specifically for generating meaningful sentence embeddings. It transforms Korean text into 768-dimensional dense vector representations, making it particularly effective for semantic similarity tasks and natural language understanding applications.
Implementation Details
The model utilizes a BERT-based architecture with mean pooling and is trained using MultipleNegativesRankingLoss with a scale factor of 20.0. It employs the AdamW optimizer with a learning rate of 2e-05 and implements a WarmupLinear scheduler with 889 warmup steps.
- Maximum sequence length: 128 tokens
- Pooling strategy: Mean tokens pooling
- Training dataset: KorNLI
- Evaluation dataset: KorSTS
Core Capabilities
- Sentence embedding generation for Korean text
- Semantic similarity computation
- Cross-sentence relationship understanding
- Support for clustering and semantic search applications
Frequently Asked Questions
Q: What makes this model unique?
This model is specifically optimized for Korean language understanding, trained on KorNLI dataset and evaluated on KorSTS, achieving impressive performance metrics including 82.24% Cosine Pearson correlation.
Q: What are the recommended use cases?
The model excels in tasks such as semantic similarity comparison, document clustering, information retrieval, and semantic search applications specifically for Korean language content.
