Views
No views yet
| Model | Hits@1 | Hits@5 | MRR | Params |
|---|---|---|---|---|
| KazEmbed-V5 (Ours) | 72% | 96% | 0.835 | 278M |
| multilingual-e5-base | 72% | 96% | 0.818 | 278M |
| multilingual-e5-large | 85% | 99% | 0.909 | 560M |
| paraphrase-mpnet-v2 | 53% | 80% | 0.648 | 278M |
| LaBSE | 48% | 73% | 0.601 | 471M |
pip install sentence-transformers1from sentence_transformers import SentenceTransformer
2
3model = SentenceTransformer('YOUR_USERNAME/kazembed-v5')
4
5# For queries (questions)
6query = "query: Қазақстанның астанасы қай қала?"
7query_embedding = model.encode(query)
8
9# For passages (documents)
10passage = "passage: Астана — Қазақстан Республикасының астанасы."
11passage_embedding = model.encode(passage)
12
13# Calculate similarity
14from sklearn.metrics.pairwise import cosine_similarity
15similarity = cosine_similarity([query_embedding], [passage_embedding])[0][0]
16print(f"Similarity: {similarity:.4f}")1from sentence_transformers import SentenceTransformer
2import numpy as np
3
4model = SentenceTransformer('YOUR_USERNAME/kazembed-v5')
5
6# Your document corpus
7documents = [
8 "Астана — Қазақстан Республикасының астанасы.",
9 "Алматы — Қазақстанның ең үлкен қаласы.",
10 "Қазақстан — Орталық Азиядағы мемлекет.",
11]
12
13# Encode documents (do once, store in vector DB)
14doc_embeddings = model.encode(["passage: " + doc for doc in documents])
15
16# Query
17query = "Қазақстанның астанасы қай қала?"
18query_embedding = model.encode("query: " + query)
19
20# Find most similar
21similarities = np.dot(doc_embeddings, query_embedding)
22best_idx = np.argmax(similarities)
23print(f"Best match: {documents[best_idx]}")| Dataset | Pairs | Description |
|---|---|---|
| KazQAD | 6,640 | Question-Context pairs |
| KazQAD-Retrieval | 44,615 | Title-Text pairs |
| Powerful-Kazakh-Dialogue | 10,000 | User-Assistant pairs |
| Total | 61,255 | Retrieval-focused pairs |
query: and passage: prefixes for best results1@misc{kazembed2024,
2 title={KazEmbed-V5: A Fine-tuned Embedding Model for Kazakh Language Retrieval},
3 author={Your Name},
4 year={2024},
5 howpublished={HuggingFace Hub}
6}