paraphrase-multilingual-MiniLM-L12-v2

sentence-transformers Apache 2.0 out 5 years ago

118M parameters, text in, embeddings out, 512 tokens of context. 1 provider serves it.

Verified 2026-10-25 Rank #4 in Embeddings · score 111.7 method v1 Sources 3 Downloads 30d 40.4M · #2 of 619 tracked

Changed this week: 4 new quantized builds on the Hub (38 in all).2026-10-24 · 3 new quantized builds on the Hub (37 in all).2026-10-24

Can I run it?0.1GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24

Does it work?56.42on MTEB#36 of 64 measuredleaderboard2026-10-24

Do people use it?40.4Mdownloads, 30 days#2 of 619 we track · top 1%+12 likes, downloads −12% this week1,489 likesHugging Face2026-10-24

Worth reading

Picked by rule from 4 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Latest · 6 months ago

paraphrase-multilingual-MiniLM-L12-v2 — Model 56/100 | Unfragile

unfragile.ai 2026-05-03 58 words

4 best of 4 tracked

The 2 picks above, plus these 2. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

Articleparaphrase-multilingual-MiniLM-L12-v2 — Use Cases, Pros & Cons | AI Tools Directoryaiappdex.com · undated
Articleparaphrase-multilingual-MiniLM-L12-v2 — Benchmarks, Pricing & Downloads - Databubbledatabubble.co · undated

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

#	Cheapest live row	In /1M	Out /1M	Context	Quant	Seen
1	HF Inference	—	—	—	—	Oct 25

All 1 provider →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Served by 1 API provider; 38 quantized builds for local use; a big base for builders (350 fine-tunes, adapters and merges).

As a service1HF Inference
On own hardware38quantized builds
Built on350350 fine-tunes · 0 adapters · 0 merges
In apps100+Hugging Face Spaces

Top Spaces running it: embedl/hfviewer ♥ 181 · yenniejun/tokenizers-languages ♥ 34 · derek-thomas/arabic-RAG ♥ 19 · Waflon/mistral-PDF-chat ♥ 9

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 0.1 GB at INT4, 0.1 GB at FP8, 0.3 GB at BF16: weights only, so leave room for the KV cache.

Estimate params × bytes per param × 1.2 · how we estimate

How it scores

Third-party measurements only; author-reported numbers stay in Reference, marked as such.

#36 of 64 on MTEB so far

56.42. Each tick is a model on this site with a measured MTEB row; author-reported numbers stay in Reference. Mean(task) x100 over 41 tasks. 41 tasks, split=test

Source · leaderboard · 2026-10-24

The story so far

Weights, every dated article, and repo commits on one rail.

1 dated article and 25 repo commits so far

Teal is the weights (2022-03-02), amber triangles are 1 dated article (each links to it), grey ticks are repo commits.

What people built on it

388 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

388 derivatives on the Hub, most of them fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF mykor/paraphrase-multilingual-MiniLM-L12-v2.gguf for llama.cpp, Ollama, LM Studio 15.5k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

3 rows

Benchmarks 3 rows

measured third-party leaderboard row (3)author-reported number from the model card or eval filesNo confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
measured MTEB · Embed text for search	35.93 nDCG@10 x100 (mean of 10 tasks)	10 tasks, split=test	leaderboard	2026-10-24
measured MTEB · Embed text for search	56.42 mean(task) x100 over 41 tasks	41 tasks, split=test	leaderboard	2026-10-24
measured MTEB Reranking · Rerank search results	45.43 MAP x100 (mean of 2 tasks)	2 tasks, split=test	leaderboard	2026-10-24

What the card says

1 cited sentence

Summary generated, cited

It maps sentences and paragraphs to a 384 dimensional dense vector space for applications like clustering or semantic search.^[1]

“maps sentences & paragraphs to a 384 dimensional dense ve…”

Generated 2026-10-07 14:59 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

3 use cases

What it is good at 3 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

21 fields

Listed in Embeddings

Facts

paraphrase-multilingual-MiniLM-L12-v2 has 118M parameters. (safetensors metadata, observed 2026-10-24)
Context length is 512 tokens (512). (config.json max_position_embeddings, observed 2026-10-24)
License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
Created 2022-03-02, last modified 2026-01-28. (HF API, observed 2026-10-24)

Pipeline tag	sentence-similarity
Library	sentence-transformers
Architectures	BertModel
Model type	bert
Params by dtype	F32: 117,653,760 · I64: 512
Size band	S
MoE	no
Context length	512 (config.json max_position_embeddings)
Languages	multilingual, ar, bg, ca, cs, da, de, el, en, es, et, fa, fi, fr, gl, gu, he, hi, hr, hu, hy, id, it, ja, ka, ko, ku, lt, lv, mk, mn, mr, ms, my, nb, nl, pl, pt, ro, ru, sk, sl, sq, sr, sv, th, tr, uk, ur, vi
Papers	arXiv:1908.10084
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · BertModel · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · BertModel · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for paraphrase-multilingual-MiniLM-L12-v2: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	0.3 GB · Edge (<4 GB)
FP8 (1 byte/param)	0.1 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)	0.1 GB · Edge (<4 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Original model Family: sentence-transformers/paraphrase 388 derivatives on the Hub

quantized: 38finetune: 350

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
Xenova/paraphrase-multilingual-MiniLM-L12-v2	quantized	—	—	—	—	Xenova	19	15.2M
gmunkhtur/paraphrase-multilingual-minilm-l12-v3-mn	finetune	—	118M	S	—	gmunkhtur	0	545.7k
mykor/paraphrase-multilingual-MiniLM-L12-v2.gguf	quantized	gguf	—	—	apache-2.0	mykor	3	15.5k
Decius/botpress_Vaganet_new_model	finetune	—	118M	S	—	Decius	0	14.7k
Omartificial-Intelligence-Space/Arabic-MiniLM-L12-v2-all-nli-triplet	finetune	—	118M	S	apache-2.0	Omartificial-Intelligence-Space	4	13.5k
vishnu-n/Morgan-Tanglish-v7	finetune	—	118M	S	apache-2.0	vishnu-n	0	8.3k
faodl/20250909_model_g20_multilabel_MiniLM-L12-all-labels-artificial-governance-multi-output	finetune	—	118M	S	—	faodl	0	6.9k
gmunkhtur/paraphrase-mongolian-minilm-mn_v2	finetune	—	118M	S	—	gmunkhtur	2	6.1k
hothanhtienqb/mind_map_blog_model	finetune	—	118M	S	—	hothanhtienqb	3	4.4k
lengocquangLAB/fine-tuned-jobtitle-embed	finetune	—	118M	S	—	lengocquangLAB	0	2.7k
cstr/paraphrase-multilingual-MiniLM-L12-v2-GGUF	quantized	GGUF	—	—	apache-2.0	cstr	0	2.4k
yahyaabd/allstats-search-mini-v1-1-mnrl	finetune	—	118M	S	—	yahyaabd	0	2k
fabskill/job_and_title_siamese_binary	finetune	—	118M	S	—	fabskill	0	1.8k
Chernoffface/fs-setfit-multilable-model	finetune	—	118M	S	—	Chernoffface	1	1.6k
ciCic/paraphrase-multilingual-MiniLM-L12-v2-sts-2d-matryoshka	finetune	—	118M	S	—	ciCic	1	1.4k
tikanosa/fine-tuned-sbert-prodi	finetune	—	118M	S	—	tikanosa	0	1.2k
Lorg0n/hikka-forge-paraphrase-multilingual-MiniLM-L12-v2	quantized	—	118M	S	apache-2.0	Lorg0n	2	974
SamuelMauli/parity-embedding-juridico-br-v4	quantized	—	118M	S	apache-2.0	SamuelMauli	3	875
easonanalytica/cnm-multilingual-small-v2	finetune	—	118M	S	—	easonanalytica	0	1k
karsar/paraphrase-multilingual-MiniLM-L12-hu-v3	finetune	—	118M	S	apache-2.0	karsar	0	967
BlackKakapo/cupidon-small-ro	quantized	—	118M	S	apache-2.0	BlackKakapo	2	838
Observer2399/erdanei-security-encoder	finetune	—	118M	S	—	Observer2399	0	935
darck-12/news-classification-minilm	finetune	—	118M	S	—	darck-12	2	821
onnx-community/paraphrase-multilingual-MiniLM-L12-v2-ONNX	quantized	—	—	—	apache-2.0	onnx-community	2	805
Kelnux/seo-embedding-uczciweseo	quantized	GGUF	118M	S	mit	Kelnux	0	904
DatarrX/myX-Semantic	finetune	—	118M	S	apache-2.0	DatarrX	4	676
SMARTICT/paraphrase-multilingual-MiniLM-L12-v2-ft-tr-rag-v1	finetune	—	118M	S	apache-2.0	SMARTICT	0	759
DatarrX/myX-Semantic-Light	finetune	—	118M	S	apache-2.0	DatarrX	5	461
Whyteeeee/paraphrase-multilingual-MiniLM-L12-v2.gguf	quantized	gguf	—	—	apache-2.0	Whyteeeee	0	698
hatemestinbejaia/mmarco-Arabic-mMiniLML-bi-encoder-KD-v1	finetune	—	118M	S	—	hatemestinbejaia	0	674
ayakiri/wolo-app-categories-setfit-model	finetune	—	118M	S	—	ayakiri	0	638
tikanosa/fine-tuned-edusbert	finetune	—	118M	S	—	tikanosa	0	618
josangho99/ko-paraphrase-multilingual-MiniLM-L12-v2-multiTask	finetune	—	118M	S	—	josangho99	0	577
Geo150/addresses-tuned-for-latam	finetune	—	118M	S	—	Geo150	0	570
biblissima/all-MiniLM-L6-v2_cotes	finetune	—	118M	S	—	biblissima	1	509
saraleivam/GURU-paraphrase-multilingual-MiniLM-L12-v2	finetune	—	118M	S	—	saraleivam	1	499
DashReza7/paraphrase-multilingual-MiniLM-L12-v2_QuoraDuplicateDetection_FINETUNED	finetune	—	118M	S	—	DashReza7	0	549
Karmelkke/everycli-minilm-ft	finetune	—	118M	S	—	Karmelkke	0	548
Ahmedhisham/queen_of_embedded_egy_20k	finetune	—	118M	S	apache-2.0	Ahmedhisham	0	526
ostoveland/test7	finetune	—	—	—	—	ostoveland	0	519

Repository changelog

25 commits

changed in the last 7 daysOther: 2026-10-24, 2026-10-24all changes

Repository changelog 25 commits

Date	Commit	SHA
2026-01-28	Set tokenizer_class to PreTrainedTokenizerFast for transformers v5 compat.	e8f8c21
2025-03-06	Remove deprecated (SEB) evaluation results section	86741b4
2024-11-05	Add exported openvino model 'openvino_model_qint8_quantized.xml'	8d6b950
2024-10-11	Add new SentenceTransformer model with an openvino backend	ae06c00
2024-10-11	Add new SentenceTransformer model with an openvino backend	16e5344
2024-10-11	Add exported ONNX model 'model_qint8_avx512_vnni.onnx'	a3ded7f
2024-10-11	Add exported ONNX model 'model_qint8_avx512.onnx'	aa6fa97
2024-10-11	Add exported ONNX model 'model_quint8_avx2.onnx'	a281789
2024-10-11	Add exported ONNX model 'model_qint8_arm64.onnx'	a184365
2024-10-11	Add exported ONNX model 'model_O4.onnx'	207a9f5
2024-10-11	Add exported ONNX model 'model_O3.onnx'	5d57dee
2024-10-11	Add exported ONNX model 'model_O2.onnx'	b081807
2024-10-11	Add exported ONNX model 'model_O1.onnx'	3241ac9
2024-10-11	Add new SentenceTransformer model with an onnx backend	f16484b
2024-03-27	Adding `safetensors` variant of this model (#13)	bf3bf13
2024-02-12	Update Sentence Transformers metadata (#12)	43dcf58
2023-11-02	Add missing space preceding comment	a9c5552
2023-11-02	Update README.md (#2)	981b093
2022-06-15	Add TF weights	b8ef008
2021-11-02	Update README.md	e04d38a
2021-08-05	Update README.md	d66eff4
2021-06-23	Add new SentenceTransformer model.	e625097
2021-06-09	Create README.md	701f60b
2021-06-02	upload	92daac3
2021-06-02	initial commit	80fe43f

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for paraphrase-multilingual-MiniLM-L12-v2, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
HF Inference hf	sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

4 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Analysis4

paraphrase-multilingual-MiniLM-L12-v2 — Model 56/100 | Unfragile

unfragile.ai2026-05-0358 words

paraphrase-multilingual-MiniLM-L12-v2 by sentence-transformers - Overview, Benchmarks & Hardware | Q4KM.ai

q4km.ai967 words

The sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 model is a multilingual sentence-embedding encoder built on the MiniLM-L12 architecture that converts sentences into 384-dimensional vectors across 30+ languages. It features 12 transformer layers and requires approximately 1 GB of VRAM when using FP16, achieving inference times of roughly 2 ms on an RTX 2070. Released under the Apache 2.0 license, the model offers a favorable trade-off between speed and accuracy for semantic search and duplicate detection tasks.written, reviewed, cited

The model supports a maximum sequence length of 128 tokens using the WordPiece tokenizer from the original MiniLM-L12 checkpoint.
Training required a cluster of 8 × NVIDIA V100 GPUs for roughly 12 hours employing mixed-precision (FP16) to accelerate convergence.
Adding the Safetensors variant reduces the total model file download size by approximately 30 %.

paraphrase-multilingual-MiniLM-L12-v2 — Use Cases, Pros & Cons | AI Tools Directory

aiappdex.com368 words

The sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2 model supports over 50 languages for cross-lingual semantic search and duplicate detection. It has accumulated 45,263,676 downloads and uses 384-dim outputs to keep vector store costs low relative to 768-dim alternatives. However, the lack of instruction tuning means prompt phrasing affects embedding quality noticeably, and accuracy gaps exist between high-resource and low-resource languages.written, reviewed, cited

paraphrase-multilingual-MiniLM-L12-v2 — Benchmarks, Pricing & Downloads - Databubble

databubble.co338 words

paraphrase-multilingual-MiniLM-L12-v2 is a 59M parameter model released by SBERT registered under the sentence-similarity pipeline tag on Hugging Face. The apache-2.0 license permits commercial deployment and derivative work without per-seat fees, though attribution requirements still apply. The model is described as best fit for general-purpose chat and instruction-following workloads despite its sentence-similarity registration.written, reviewed, cited

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks cited in card

Nils Reimers, Iryna Gurevych · 2019-08-27 · PDF · HF paper page (19 upvotes)

Similar models

3 models

Similar models

multilingual-e5-small · 118M
same size band (S), same task (sentence-similarity), permissive license
all-mpnet-base-v2 · 109M
same size band (S), same task (sentence-similarity), more providers, permissive license
e5-base-v2 · 109M
same size band (S), same task (sentence-similarity), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:32 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 14:59 UTC · Last verified 2026-10-25

Follow paraphrase-multilingual-MiniLM-L12-v2

Watch this model

/feeds/models/sentence-transformers--paraphrase-multilingual-MiniLM-L12-v2.xml(RSS 2.0) · every change to paraphrase-multilingual-MiniLM-L12-v2
/feeds/families/sentence-transformers--paraphrase.xml(RSS 2.0) · every change across the paraphrase family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.