gomyk/jina-v5-student-jina_v5_compressed

jina_v5_compressed

Compact multilingual sentence encoder compressed from jinaai/jina-embeddings-v5-text-nano (11x compression).

Model Details

Property	Value
Base model	jinaai/jina-embeddings-v5-text-nano
Architecture	eurobert (decoder)
Hidden dim	384 (from 768)
Layers	6 (from 12)
Intermediate	1536
Attention heads	6
KV heads	6
Vocab size	12,796 (from 128,256)
Parameters	~19.1M
Model size (FP32)	72.8MB
Compression	11x
Distilled	No

Quick Start

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("jina_v5_compressed", trust_remote_code=True)

sentences = [
    "Hello, how are you?",
    "안녕하세요, 잘 지내세요?",
    "こんにちは、元気ですか？",
    "你好，你好吗？",
]

embeddings = model.encode(sentences)
print(embeddings.shape)  # (4, 384)

MTEB Evaluation Results

Overall Average: 38.07%

Task Group	Average
Classification	44.38%
Clustering	28.49%
STS	40.96%

Classification

Task	Average	Details
AmazonCounterfactualClassification	59.99%	en-ext: 61.54%, en: 61.13%, ja: 59.75%, de: 57.53%
Banking77Classification	45.48%	default: 45.48%
ImdbClassification	53.35%	default: 53.35%
MTOPDomainClassification	46.78%	es: 50.92%, en: 48.38%, de: 48.08%, hi: 46.87%, fr: 44.64%
MassiveIntentClassification	28.79%	en: 35.18%, zh-CN: 35.14%, lv: 34.01%, it: 33.41%, pt: 33.3%
MassiveScenarioClassification	29.45%	en: 36.67%, zh-CN: 36.26%, ms: 35.19%, sw: 34.47%, zh-TW: 33.99%
ToxicConversationsClassification	52.72%	default: 52.72%
TweetSentimentExtractionClassification	38.51%	default: 38.51%

Clustering

Task	Average	Details
ArXivHierarchicalClusteringP2P	47.49%	default: 47.49%
ArXivHierarchicalClusteringS2S	46.99%	default: 46.99%
BiorxivClusteringP2P.v2	10.67%	default: 10.67%
MedrxivClusteringP2P.v2	21.78%	default: 21.78%
MedrxivClusteringS2S.v2	20.71%	default: 20.71%
StackExchangeClustering.v2	38.42%	default: 38.42%
StackExchangeClusteringP2P.v2	32.47%	default: 32.47%
TwentyNewsgroupsClustering.v2	9.41%	default: 9.41%

STS

Task	Average	Details
BIOSSES	32.24%	default: 32.24%
SICK-R	50.98%	default: 50.98%
STS12	42.29%	default: 42.29%
STS13	47.69%	default: 47.69%
STS14	48.34%	default: 48.34%
STS15	54.33%	default: 54.33%
STS17	16.67%	es-es: 64.75%, en-en: 56.41%, ko-ko: 48.42%, ar-ar: 44.05%, es-en: 8.31%
STS22.v2	30.46%	zh: 60.89%, ar: 55.33%, fr-pl: 50.71%, es: 49.6%, it: 44.01%
STSBenchmark	45.65%	default: 45.65%

Training

Created via multi-method model compression (no additional training):

1. Teacher: jinaai/jina-embeddings-v5-text-nano (12L, 768d, 212M params)
2. Layer pruning: 12 → 6 layers (uniform selection)
3. Hidden dim: 768 → 384
4. Vocab pruning: 128,256 → 12,796 (90% cumulative frequency)
5. Compression ratio: 11x

License

This model is a derivative of Jina AI's jina-embeddings-v5-text-nano. The original model is provided under CC BY-NC 4.0 license. See jina-embeddings-v5-text-nano for details.

Supported Languages (16)

ko, en, ja, zh, es, fr, de, pt, it, ru, ar, hi, th, vi, id, pl