multilingual-e5-base-jkn-id-v1
fine-tuned model for semantic search and text matching
1 min readupdated
This specialized fine-tune of intfloat/multilingual-e5-base is built for semantic search, clustering, and text matching. It encodes sentences and paragraphs into a 768-dimensional dense vector space, making it useful for developers and researchers who need accurate document retrieval, phrase similarity analysis, or knowledge base organization.
Key features
- Built on the efficient intfloat/multilingual-e5-base architecture with support for sequence lengths up to 512 tokens.
- Outputs 768-dimensional embeddings for accurate cosine similarity calculations.
- Trained on domain-specific text data (including legal and regulatory documents) using the MultipleNegativesRankingLoss objective.
- Delivers high accuracy on information retrieval tasks.
What you can do
- Set up semantic search across internal knowledge bases, legal records, or medical documents.
- Automate text document clustering and categorization.
- Mine paraphrases and detect semantic duplicates in user queries.
- Integrate the model into NLP pipelines via the sentence-transformers library.
Example
Here is an inference example using sentence-transformers to score similarity between a query and documents:
from sentence_transformers import SentenceTransformer
model = SentenceTransformer("chrystian-yunus/multilingual-e5-base-jkn-id-v1")
queries = ['query: apakah persetujuan itu sah']
documents = ['passage: Rujukan pelayanan kesehatan...']
query_embeddings = model.encode_query(queries)
document_embeddings = model.encode_document(documents)
similarities = model.similarity(query_embeddings, document_embeddings)
print(similarities)
Read more
DetailsIR metrics: cosine_ndcg@10: 0.8187, cosine_mrr@10: 0.7593, accuracy@1: 0.6203
Embedding dimension: 768
Training dataset size: 687 examples, batch size 16, 8 epochs










Comments