Natasha/Navec hudlit_v1 Model Card
This repository provides a Sentence-Transformers version of the Natasha/Navec hudlit_v1 word embeddings model.
The underlying Navec embedding weights are unchanged. This revision adds an explicit Normalize module after StaticEmbedding, so model.encode(...) returns L2-normalized sentence embeddings by default.
Source
The original word embeddings come from the Navec project:
- Repository: https://github.com/natasha/navec
- Authors: Natasha NLP project
- License: MIT License
Navec is a compact and efficient set of Russian word embeddings trained on Russian corpora.
Usage
pip install -U sentence-transformers
from sentence_transformers import SentenceTransformer
import numpy as np
model = SentenceTransformer("BorisTM/natasha_navec_hudlit_v1_12B_500K_300d_100q")
sentences = [
"Сегодня хорошая погода.",
"На улице солнечно.",
"Команда выиграла матч.",
]
embeddings = model.encode(sentences)
print(embeddings.shape) # (3, 300)
print(np.linalg.norm(embeddings, axis=1)) # close to 1.0
similarities = model.similarity(embeddings, embeddings)
print(similarities)
Results
Results on MTEB (rus, v1.1), evaluated with normalized embeddings. Scores are percentages.
| Task | navec_hudlit_v1_12B_500K_300d_100q |
navec_news_v1_1B_250K_300d_100q |
|---|---|---|
| Mean (Task, 23 tasks) | 36.94 | 36.31 |
| Mean (Task Type) | 34.59 | 34.23 |
| CEDRClassification | 34.14 | 32.81 |
| GeoreviewClassification | 33.09 | 34.03 |
| GeoreviewClusteringP2P | 34.10 | 28.78 |
| HeadlineClassification | 55.33 | 63.47 |
| InappropriatenessClassification | 53.39 | 53.15 |
| KinopoiskClassification | 45.25 | 44.89 |
| MassiveIntentClassification | 48.54 | 43.86 |
| MassiveScenarioClassification | 55.20 | 49.88 |
| MIRACLReranking | 10.88 | 10.88 |
| MIRACLRetrievalHardNegatives.v2 | 1.75 | 1.60 |
| RiaNewsRetrievalHardNegatives.v2 | 15.43 | 23.47 |
| RuBQReranking | 38.00 | 37.71 |
| RuBQRetrieval | 5.80 | 5.09 |
| RUParaPhraserSTS | 41.38 | 41.12 |
| RuReviewsClassification | 49.35 | 48.80 |
| RuSciBenchGRNTIClassification | 43.63 | 40.54 |
| RuSciBenchGRNTIClusteringP2P | 40.94 | 38.43 |
| RuSciBenchOECDClassification | 35.62 | 32.70 |
| RuSciBenchOECDClusteringP2P | 36.89 | 33.98 |
| SensitiveTopicsClassification | 19.49 | 18.57 |
| STS22 | 50.20 | 51.57 |
| TERRa | 52.57 | 53.99 |
| RuSTSBenchmarkSTS | 48.59 | 45.84 |
Evaluation artifacts for this update are stored locally in the article workspace under data/metrics/navec_baselines_mteb_rus_v1_1_summary.csv and data/metrics/navec_baselines_mteb_rus_v1_1_task_scores.csv.
License
MIT
Contact
tg: @btmalov