modernbert_L6_uniform

Lightweight sentence encoder created from answerdotai/ModernBERT-base via layer pruning + vocabulary pruning.

Model Details

Property	Value
Teacher	answerdotai/ModernBERT-base
Architecture	ModernBERT (pruned)
Hidden dim	768
Layers	6 / 22
Layer indices	[0, 4, 8, 13, 17, 21]
Strategy	6 layers, evenly spaced from ModernBERT (22L)
Parameters	63,870,720
Model size (FP32)	176.0MB
Distilled	No

Architecture

==============================================================
  TEACHER: ModernBERT  →  STUDENT: 6L / 27,279 vocab
==============================================================

            TEACHER                        STUDENT          
  ───────────────────────────    ───────────────────────────

  ┌─────────────────────────┐    ┌─────────────────────────┐
  │   Input Tokens          │    │   Input Tokens          │
  └────────────┬────────────┘    └────────────┬────────────┘
               │                              │
  ┌────────────┴────────────┐    ┌────────────┴────────────┐
  │  Embeddings             │    │  Embeddings (pruned)    │
  │  vocab:  50,368         │    │  vocab:  27,279         │
  │  dim:  768              │    │  dim:  768              │
  └────────────┬────────────┘    └────────────┬────────────┘
               │                              │
  ┌─────────────────────────┐    ┌─────────────────────────┐
  │  Layer  0               │ ──►  │  Layer  0 ← L0         │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer  1               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer  2               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer  3               │  ╳   │                         │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer  4               │ ──►  │  Layer  1 ← L4         │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer  5               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer  6               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer  7               │  ╳   │                         │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer  8               │ ──►  │  Layer  2 ← L8         │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer  9               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer 10               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer 11               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer 12               │  ╳   │                         │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer 13               │ ──►  │  Layer  3 ← L13        │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer 14               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer 15               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer 16               │  ╳   │                         │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer 17               │ ──►  │  Layer  4 ← L17        │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer 18               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer 19               │  ╳   │                         │
  ├ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─┤    │                         │
  │  Layer 20               │  ╳   │                         │
  ├─────────────────────────┤    ├─────────────────────────┤
  │  Layer 21               │ ──►  │  Layer  5 ← L21        │
  └────────────┬────────────┘    └────────────┬────────────┘
               │                              │
  ┌────────────┴────────────┐    ┌────────────┴────────────┐
  │  Mean Pooling           │    │  Mean Pooling           │
  │  → 768d embedding       │    │  → 768d embedding       │
  └─────────────────────────┘    └─────────────────────────┘

  Size: 495.8MB (FP32)           →  176.0MB (FP32)
  Params: 129,980,160        →  46,138,368
  Reduction: 64.5%
==============================================================

Quick Start

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("modernbert_L6_uniform", trust_remote_code=True)

sentences = [
    "Hello, how are you?",
    "안녕하세요",
    "Bonjour, comment allez-vous?",
]

embeddings = model.encode(sentences)
print(embeddings.shape)  # (3, 768)

MTEB Evaluation Results

Overall Average: 35.42%

Task Group	Average
Classification	41.97%
Clustering	28.21%
STS	36.0%

Classification

Task	Average	Details
AmazonCounterfactualClassification	59.33%	en-ext: 61.33%, de: 60.17%, en: 58.99%
Banking77Classification	35.01%	default: 35.01%
ImdbClassification	55.05%	default: 55.05%
MTOPDomainClassification	43.24%	en: 47.72%, es: 47.27%, th: 45.01%
MassiveIntentClassification	25.86%	zh-CN: 37.2%, ja: 34.36%, zh-TW: 32.51%
MassiveScenarioClassification	26.28%	zh-CN: 38.18%, zh-TW: 33.36%, en: 32.44%
ToxicConversationsClassification	52.6%	default: 52.6%
TweetSentimentExtractionClassification	38.42%	default: 38.42%

Clustering

Task	Average	Details
ArXivHierarchicalClusteringP2P	50.19%	default: 50.19%
ArXivHierarchicalClusteringS2S	46.96%	default: 46.96%
BiorxivClusteringP2P.v2	12.62%	default: 12.62%
MedrxivClusteringP2P.v2	22.13%	default: 22.13%
MedrxivClusteringS2S.v2	19.43%	default: 19.43%
StackExchangeClustering.v2	34.26%	default: 34.26%
StackExchangeClusteringP2P.v2	31.01%	default: 31.01%
TwentyNewsgroupsClustering.v2	9.11%	default: 9.11%

STS

Task	Average	Details
BIOSSES	33.84%	default: 33.84%
SICK-R	46.99%	default: 46.99%
STS12	35.32%	default: 35.32%
STS13	33.7%	default: 33.7%
STS14	37.07%	default: 37.07%
STS15	49.85%	default: 49.85%
STS17	23.34%	es-es: 61.45%, en-en: 55.74%, ko-ko: 48.68%
STS22.v2	24.05%	zh: 52.16%, es: 46.52%, it: 45.35%
STSBenchmark	39.82%	default: 39.82%

Training

Created via layer pruning + vocabulary pruning (no additional training):

Teacher: answerdotai/ModernBERT-base (22 layers, 768d)
Layer selection: [0, 4, 8, 13, 17, 21] - 6 layers, evenly spaced from ModernBERT (22L)
Vocab pruning: Corpus-based filtering for target languages

Supported Languages (18)

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

Downloads last month: 1

Safetensors

Model size

46.1M params

Tensor type

F32