all-mpnet-base-v2

sentence-transformers Apache 2.0 out 5 years ago

109M parameters, text in, embeddings out, 514 tokens of context. 2 providers serve it from $0.005 per million input tokens.

Verified 2026-10-25 Rank #3 in Embeddings · score 113.3 method v1 Sources 3 Downloads 30d 18.3M · #6 of 619 tracked

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

What does it cost?$0.005/1M incheapest of 1DeepInfra2026-10-24

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

Does it work?48.41on MTEB Reranking#4 of 33 measuredleaderboard2026-10-24

Do people use it?18.3Mdownloads, 30 days#6 of 619 we track · top 1%+5 likes, downloads +1% this week1,400 likesHugging Face2026-10-24

Worth reading

Picked by rule from 5 tracked pieces: the lab's own post, the most shared, a field report, the latest.

5 best of 5 tracked

The 1 pick above, plus these 4. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

Articleall-mpnet-base-v2: higher quality | Sentence Transformers Beginner Course | The Neural Basetheneuralbase.com · undated
ArticleAll Mpnet Base V2 API - Demo - DeepInfradeepinfra.com · undated
ArticleAll-Mpnet-Base-V2: Enhancing Sentence Embedding with AIzilliz.com · undated
Articleall-mpnet-base-v2 | VIPS Learnlearn.engineering.vips.edu · 2026-04-20

Where to run it

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

#	Cheapest 2	In /1M	Out /1M	Context	Quant	Seen
1	DeepInfra	$0.005	—	512	—	Oct 24
2	HF Inference	—	—	—	—	Oct 25

All 2 providers →

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 2 API providers; 8 quantized builds for local use; a big base for builders (408 fine-tunes, adapters and merges).

As a service2DeepInfra, HF Inference
On own hardware8quantized builds
Built on408407 fine-tunes · 1 adapters · 0 merges
In apps100+Hugging Face Spaces

Top Spaces running it: embedl/hfviewer ♥ 181 · sentence-transformers/embeddings-semantic-search ♥ 118 · huggingface/data-measurements-tool ♥ 100 · nickmuchi/semantic-search-with-retrieve-and-rerank ♥ 71

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.

#4 of 33 on MTEB Reranking so far

48.41. Each tick is a model on this site with a measured MTEB Reranking row; author-reported numbers stay in Reference. MAP x100 (mean of 2 tasks). 2 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 29 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

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

416 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 cstr/all-mpnet-base-v2-GGUF for llama.cpp, Ollama, LM Studio 1.1k 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	44.54 nDCG@10 x100 (mean of 10 tasks)	10 tasks, split=test	leaderboard	2026-10-24
measured MTEB · Embed text for search	59.62 mean over 39/41 tasks present	39/41 tasks	leaderboard	2026-10-24
measured MTEB Reranking · Rerank search results	48.41 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 768 dimensional dense vector space for tasks such as clustering or semantic search.^[1]

“maps sentences & paragraphs to a 768 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

2 use cases

What it is good at 2 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

all-mpnet-base-v2 has 109M parameters. (safetensors metadata, observed 2026-10-24)
Context length is 514 tokens (514). (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 2025-08-19. (HF API, observed 2026-10-24)

Pipeline tag	sentence-similarity
Library	sentence-transformers
Architectures	MPNetForMaskedLM
Model type	mpnet
Params by dtype	F32: 109,486,464 · I64: 514
Size band	S
MoE	no
Context length	514 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:1904.06472, arXiv:2102.07033, arXiv:2104.08727, arXiv:1704.05179, arXiv:1810.09305
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: not in registry · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 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 all-mpnet-base-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/all-mpnet-v2 416 derivatives on the Hub

finetune: 407quantized: 8adapter: 1

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
TechWolf/JobBERT-v2	finetune	—	109M	S	mit	TechWolf	40	1.7M
farbodtavakkoli/OTel-Embedding-109M	finetune	—	109M	S	apache-2.0	farbodtavakkoli	1	1.2M
Xenova/all-mpnet-base-v2	quantized	—	—	—	—	Xenova	6	855.2k
s2593817/sft-sql-embedding	finetune	—	109M	S	—	s2593817	4	307.8k
lealdaniel/comp-embedding-matching	finetune	—	109M	S	—	lealdaniel	0	94k
TechWolf/ConTeXT-Skill-Extraction-base	finetune	—	109M	S	—	TechWolf	8	80k
dwulff/mpnet-personality	finetune	—	109M	S	cc-by-sa-4.0	dwulff	16	40.4k
jonny9f/food_embeddings	finetune	—	109M	S	—	jonny9f	0	20k
candidatePI/candidate-ft-model	finetune	—	109M	S	—	candidatePI	0	16.3k
mihirsingh141/retriever_module	finetune	—	109M	S	apache-2.0	mihirsingh141	0	10.8k
sindsub/cybersec-mpnet-finetuned	finetune	—	109M	S	apache-2.0	sindsub	1	10.4k
vida-nyu/magneto-schema-retriever-gdc	finetune	—	109M	S	apache-2.0	vida-nyu	2	9.4k
wwydmanski/all-mpnet-base-v2-legal-v0.1	finetune	—	109M	S	—	wwydmanski	1	6.1k
Aleksandruz/skillmatch-mpnet-curriculum-retriever	finetune	—	109M	S	mit	Aleksandruz	2	5.5k
MasterControlAIML/finetuned-ecfr-embeddings	finetune	—	109M	S	—	MasterControlAIML	0	3.9k
ElenaSenger/career-path-representation-mpnet-decorte	finetune	—	109M	S	apache-2.0	ElenaSenger	2	3.1k
magnolia-psychometrics/surveybot3000	finetune	—	109M	S	cc-by-nc-4.0	magnolia-psychometrics	5	3k
shankerram3/resumator	finetune	—	109M	S	apache-2.0	shankerram3	1	1.8k
citiusLTL/bdi-all-mpnet-base-v2	finetune	—	109M	S	apache-2.0	citiusLTL	0	1.7k
harmonydata/mental_health_harmonisation_1	finetune	—	109M	S	—	harmonydata	2	1.6k
richie-ghost/sentence-transformers-all-mpnet-base-v2	finetune	—	109M	S	—	richie-ghost	0	1.7k
Romaniox/all-mpnet-base-v2-sts	finetune	—	109M	S	—	Romaniox	0	1.5k
alpcansoydas/product-model-16.10.24	finetune	—	109M	S	—	alpcansoydas	0	1.5k
enochlev/coherence-all-mpnet-base-v2	finetune	—	109M	S	—	enochlev	3	1.1k
jonny9f/food_embeddings2	finetune	—	109M	S	—	jonny9f	1	1.2k
ElenaSenger/career-path-representation-mpnet-karrierewege	finetune	—	109M	S	apache-2.0	ElenaSenger	1	1.2k
JJTsao/fine-tuned_movie_retriever-all-mpnet-base-v2	finetune	—	109M	S	apache-2.0	JJTsao	0	1.2k
sarthak1/codemalt	finetune	—	7.56M	S	apache-2.0	sarthak1	2	1.1k
Shobhank-iiitdwd/Clinical_sentence_transformers_mpnet_base_v2	finetune	—	109M	S	—	Shobhank-iiitdwd	1	1.1k
cstr/all-mpnet-base-v2-GGUF	quantized	GGUF	—	—	apache-2.0	cstr	0	1.1k
nafis277/domain-mpnet-normalized	finetune	—	109M	S	—	nafis277	0	1.1k
Prashasst/anime-recommendation-model	finetune	—	109M	S	—	Prashasst	2	948
joshuasundance/setfit-absa-all-mpnet-base-v2-laptops-polarity	finetune	—	109M	S	—	joshuasundance	1	951
bwang0911/jev2-legal	finetune	—	109M	S	—	bwang0911	1	909
marcelomoreno26/all-mpnet-base-v2-absa-polarity	finetune	—	109M	S	—	marcelomoreno26	0	922
hassaangyt/compliance-emb-v1	finetune	—	109M	S	—	hassaangyt	0	922
CrosswaveOmega/DMRSQ-Utterance-To-Item-Transformer	finetune	—	109M	S	—	CrosswaveOmega	0	908
hojzas/proj4-all-labs	finetune	—	109M	S	—	hojzas	0	903
Hierarchy-Transformers/HiT-MPNet-WordNetNoun	finetune	—	109M	S	apache-2.0	Hierarchy-Transformers	0	761
mtyrrell/CPU_Transport_GHG_Classifier	finetune	—	—	—	apache-2.0	mtyrrell	0	741

Repository changelog

29 commits

Repository changelog 29 commits

Date	Commit	SHA
2025-08-19	Add `text-embeddings-inference` tag & snippet (#40)	e8c3b32
2025-03-06	Remove deprecated (SEB) evaluation results section	12e86a3
2024-11-05	Add exported openvino model 'openvino_model_qint8_quantized.xml'	9a32259
2024-10-11	Add new SentenceTransformer model with an openvino backend	f1b1b82
2024-10-11	Add new SentenceTransformer model with an openvino backend	a6650fe
2024-10-11	Add exported ONNX model 'model_qint8_avx512_vnni.onnx'	84f651b
2024-10-11	Add exported ONNX model 'model_qint8_avx512.onnx'	5eb55be
2024-10-11	Add exported ONNX model 'model_quint8_avx2.onnx'	45a099c
2024-10-11	Add exported ONNX model 'model_qint8_arm64.onnx'	687ca85
2024-10-11	Add exported ONNX model 'model_O4.onnx'	ceafd97
2024-10-11	Add exported ONNX model 'model_O3.onnx'	edf997a
2024-10-11	Add exported ONNX model 'model_O2.onnx'	4fb5900
2024-10-11	Add exported ONNX model 'model_O1.onnx'	14445e2
2024-10-11	Add new SentenceTransformer model with an onnx backend	21eca97
2024-03-27	Adding `safetensors` variant of this model (#19)	84f2bcc
2024-02-12	Update Sentence Transformers metadata (#16)	c46f31f
2023-11-02	Resolve broken metadata tag (#12)	5681fe0
2022-07-11	adding training dataset	bd44305
2022-07-08	adding training dataset	baf49b6
2022-07-08	Update README.md	721a244
2022-07-08	adding training datasets	2f25f07
2022-07-06	adding training datasets	25c9cf2
2022-07-06	adding training datasets	483eba7
2022-07-02	Add s2orc dataset	cd48db1
2021-10-15	Update README.md	86eb780
2021-10-15	Update README.md	955d069
2021-08-31	up	41a945f
2021-08-30	upload	58ae313
2021-08-30	initial commit	d58b8e2

All providers

2 providers · 2 endpoints

Where to run 2 providers · 2 endpoints

Hosted endpoints observed for all-mpnet-base-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. Cheapest observed input price: $0.0050/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
DeepInfra deepinfra	sentence-transformers/all-mpnet-base-v2	512	—	$0.0050	—	—	—	—	live	2026-10-24
HF Inference hf	sentence-transformers/all-mpnet-base-v2	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

5 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.

Analysis5

all-mpnet-base-v2 by sentence-transformers - Overview, Benchmarks & Hardware | Q4KM.ai

q4km.ai1,125 words

The all-mpnet-base-v2 model converts English sentences into 768-dimensional vectors for semantic search and clustering. Built on a 12-layer MPNet transformer with ~110 M parameters, it was trained on a 1 B-sentence-pair corpus using contrastive learning. The Apache-2.0 license allows commercial use, while CPU inference runs at ~200-300 ms per sentence on a 12-core Xeon.written, reviewed, cited

The model requires a GPU with at least 6 GB VRAM to handle batch sizes of 32 to 64 sentences.
Training utilized 8 NVIDIA V100 GPUs running mixed-precision for approximately 2 weeks.
The model can be further fine-tuned on task-specific data without altering the underlying MPNet weights.

all-mpnet-base-v2: higher quality | Sentence Transformers Beginner Course | The Neural Base

theneuralbase.com794 words

all-mpnet-base-v2 is a 110M parameter transformer model fine-tuned on over 1 billion sentence pairs that produces 768-dimensional embeddings for high-precision semantic search. The 438MB model achieves 15 min encoding for 1M documents at batch_size=128 only on an A100 GPU, while upgrading the sentence-transformers library can cause version-induced embedding drift that breaks similarity searches. Although all-MiniLM-L6-v2 is 5x faster, it suffers a 38% accuracy loss on legal and medical tasks compared to all-mpnet-base-v2.written, reviewed, cited

Embedding outputs vary across sentencetransformers versions due to floatingpoint precision and tokenizer updates.
CPU encoding for 1M documents takes 45+ minutes and is not recommended for production workloads.
Cold start latency on AWS Lambda with ephemeral storage reaches 34 seconds before the model loads.

All Mpnet Base V2 API - Demo - DeepInfra

deepinfra.com981 words

DeepInfra hosts the All-MPNet-Base-V2 model for $0.005 per million tokens. This sentence-transformers model maps text to a 768 dimensional dense vector space and was fine-tuned on a 1B sentence pairs dataset. The total number of training tuples across all datasets reaches 1,170,060,424.written, reviewed, cited

Input text longer than 384 word pieces is truncated by default.
The model uses a 2e-5 learning rate with an AdamW optimizer.
Requests on the 'flex' tier may be retried or timed out under load.

All-Mpnet-Base-V2: Enhancing Sentence Embedding with AI

zilliz.com3,279 words

All-Mpnet-Base-V2 is a sentence embedding model designed to handle long-form text by combining the strengths of BERT and XLNet. The model captures bidirectional and autoregressive information by fusing Masked Language Modeling and Permuted Language Modeling in its operation. This unified approach allows the model to excel in semantic search tasks where understanding nuanced sentence meaning is paramount.written, reviewed, cited

The model utilizes two-stream self-attention to allow each token to consider previously predicted tokens during training.
MPNet incorporates the mask symbol and positional information of masked tokens to ensure tokens see the full input context.
The model maps sentences to dense vector spaces, enabling effective clustering and semantic search for retrieval tasks.

all-mpnet-base-v2 | VIPS Learn

learn.engineering.vips.edu2026-04-20280 words

all-mpnet-base-v2 is a widely deployed open-weights text-embedding model published in 2021 by the sentence-transformers community. At 110M parameters it fits comfortably on CPU and offers Apache 2.0-compatible open weights via sentence-transformers. The model supports a 384-token context window and scores an MTEB English average of ≈58 as of 2023-03.written, reviewed, cited

Papers

5 papers

Papers 5

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).

A Repository of Conversational Datasets cited in card peer-reviewed · reported

Matthew Henderson, Paweł Budzianowski, Iñigo Casanueva +8 · 2019-04-13 · Proceedings of the Workshop on NLP for Conversational AI (2019) · PDF · HF paper page (5 upvotes) · evidence
GooAQ: Open Question Answering with Diverse Answer Types cited in card peer-reviewed · measured

Daniel Khashabi, Amos Ng, Tushar Khot +3 · 2021-04-18 · Empirical Methods in Natural Language Processing · PDF · HF paper page (1 upvotes) · evidence
PAQ: 65 Million Probably-Asked Questions and What You Can Do With Them cited in card peer-reviewed · measured

Patrick Lewis, Yuxiang Wu, Linqing Liu +5 · 2021-02-13 · Transactions of the Association for Computational Linguistics · PDF · HF paper page · evidence
SearchQA: A New Q&A Dataset Augmented with Context from a Search Engine cited in card

Matthew Dunn, Levent Sagun, Mike Higgins +3 · 2017-04-18 · PDF · HF paper page (1 upvotes)
WikiHow: A Large Scale Text Summarization Dataset cited in card

Mahnaz Koupaee, William Yang Wang · 2018-10-18 · PDF · HF paper page

Similar models

3 models

Similar models

paraphrase-multilingual-MiniLM-L12-v2 · 118M
same size band (S), same task (sentence-similarity), fewer providers, permissive license
e5-base-v2 · 109M
same size band (S), same task (sentence-similarity), permissive license
multi-qa-mpnet-base-dot-v1 · 109M
same size band (S), same task (sentence-similarity)

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 all-mpnet-base-v2

Watch this model

/feeds/models/sentence-transformers--all-mpnet-base-v2.xml(RSS 2.0) · every change to all-mpnet-base-v2
/feeds/families/sentence-transformers--all-mpnet-v2.xml(RSS 2.0) · every change across the all-mpnet-v2 family

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