gomyk/me5s-student-me5s_compressed

me5s_compressed

Compact multilingual sentence encoder compressed from intfloat/multilingual-e5-small (9x compression).

Model Details

Property	Value
Base model	intfloat/multilingual-e5-small
Architecture	bert (encoder)
Hidden dim	384 (from 384)
Layers	4 (from 12)
Intermediate	1536
Attention heads	12
Vocab size	15,168 (from 250,037)
Parameters	~13.1M
Model size (FP32)	50.6MB
Compression	9x
Distilled	No

Quick Start

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("me5s_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: 46.19%

Task Group	Average
Classification	52.2%
Clustering	30.4%
STS	54.88%

Classification

Task	Average	Details
AmazonCounterfactualClassification	67.37%	en: 69.31%, en-ext: 67.39%, ja: 67.39%, de: 65.39%
Banking77Classification	58.7%	default: 58.7%
ImdbClassification	57.14%	default: 57.14%
MTOPDomainClassification	66.84%	en: 75.99%, es: 69.48%, hi: 68.15%, fr: 63.63%, th: 63.38%
MassiveIntentClassification	31.12%	en: 53.03%, zh-CN: 51.62%, it: 47.56%, pt: 47.28%, ja: 47.03%
MassiveScenarioClassification	34.85%	zh-CN: 59.05%, en: 58.06%, ja: 51.79%, it: 50.05%, vi: 49.68%
ToxicConversationsClassification	55.82%	default: 55.82%
TweetSentimentExtractionClassification	45.74%	default: 45.74%

Clustering

Task	Average	Details
ArXivHierarchicalClusteringP2P	47.08%	default: 47.08%
ArXivHierarchicalClusteringS2S	48.29%	default: 48.29%
BiorxivClusteringP2P.v2	17.24%	default: 17.24%
MedrxivClusteringP2P.v2	24.42%	default: 24.42%
MedrxivClusteringS2S.v2	21.55%	default: 21.55%
StackExchangeClustering.v2	39.42%	default: 39.42%
StackExchangeClusteringP2P.v2	31.85%	default: 31.85%
TwentyNewsgroupsClustering.v2	13.35%	default: 13.35%

STS

Task	Average	Details
BIOSSES	56.68%	default: 56.68%
SICK-R	59.22%	default: 59.22%
STS12	52.11%	default: 52.11%
STS13	64.25%	default: 64.25%
STS14	60.12%	default: 60.12%
STS15	74.19%	default: 74.19%
STS17	38.71%	es-es: 74.88%, en-en: 74.6%, ar-ar: 63.98%, ko-ko: 58.54%, nl-en: 28.91%
STS22.v2	27.7%	zh: 59.69%, fr: 57.76%, es: 56.19%, en: 50.93%, it: 50.09%
STSBenchmark	60.91%	default: 60.91%

Training

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

1. Teacher: intfloat/multilingual-e5-small (12L, 384d, 117M params)
2. Layer pruning: 12 → 4 layers (uniform selection)
3. Hidden dim: 384 → 384
4. Vocab pruning: 250,037 → 15,168 (90% cumulative frequency)
5. Compression ratio: 9x

Supported Languages (18)

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