Field report
github.com 38 words
The story so far
Weights, every dated article, and repo commits on one rail.
21 repo commits since the weights landed
Teal is the weights (2022-03-02), grey ticks are repo commits.
clip-ViT-B-32-multilingual-v1
sentence-transformers Apache 2.0 out 5 years ago
135M parameters, text in, embeddings out, 512 tokens of context. 2 providers serve it from $0.005 per million input tokens.
Verified 2026-10-25 Rank #52 in Embeddings · score 88.8 method v1 Sources 3 Downloads 30d 110.4k · #311 of 619 tracked
Do people use it?110.4kdownloads, 30 days#311 of 619 we track · top 50%+1 likes, downloads +3% this week195 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.
4 best of 4 tracked
The 1 pick above, plus these 3. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
- Articleclip-ViT-B-32-multilingual-v1 | PromptLayer Modelspromptlayer.com · undated
- ArticleClip ViT B 32 Multilingual V1 · Models · Dataloopdataloop.ai · undated
- ArticleClip ViT B 32 Multilingual V1 API - Voice - DeepInfradeepinfra.com · undated
Where to run it
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
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; heavily run locally (98 quantized builds).
- As a service2DeepInfra, HF Inference
- On own hardware98quantized builds
- Built on21 fine-tunes · 0 adapters · 1 merges
- In apps20Hugging Face Spaces
Top Spaces running it: merve/multilingual-zero-shot-image-clf ♥ 13 · NeuML/imagesearch ♥ 5 · svjack/context-dialogue-chinese-sample-search ♥ 1
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, 0.2 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
What people built on it
100 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
100 derivatives on the Hub, most of them quantized
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Reference
Everything else, folded: the full tables, every source.
What the card says
1 cited sentence
Summary generated, cited
It supports multilingual zero-shot image classification and image search across 50+ languages.[1]
Generated 2026-10-07 15:04 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
- clip-ViT-B-32-multilingual-v1 has 135M 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 2024-11-05. (HF API, observed 2026-10-24)
| Pipeline tag | sentence-similarity |
|---|---|
| Library | sentence-transformers |
| Architectures | DistilBertModel |
| Model type | distilbert |
| Params by dtype | F32: 134,734,080 |
| Size band | S |
| MoE | no |
| Context length | 512 (config.json max_position_embeddings) |
| Languages | multilingual |
| Papers | arXiv:2004.09813, arXiv:1908.10084 |
| 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 clip-ViT-B-32-multilingual-v1: 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) | 0.2 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
quantized: 98merge: 1finetune: 1
Top variants (40 shown, ranked by downloads)
Repository changelog
21 commits
Repository changelog 21 commits
| Date | Commit | SHA |
|---|---|---|
| 2024-11-05 | Add exported openvino model 'openvino_model_qint8_quantized.xml' | 58edf8c |
| 2024-10-11 | Add new SentenceTransformer model with an openvino backend | 04bd010 |
| 2024-10-11 | Add new SentenceTransformer model with an openvino backend | 29716be |
| 2024-10-11 | Add exported ONNX model 'model_qint8_avx512_vnni.onnx' | e5ec844 |
| 2024-10-11 | Add exported ONNX model 'model_qint8_avx512.onnx' | 2437c7b |
| 2024-10-11 | Add exported ONNX model 'model_quint8_avx2.onnx' | d498c72 |
| 2024-10-11 | Add exported ONNX model 'model_qint8_arm64.onnx' | aa2e374 |
| 2024-10-11 | Add exported ONNX model 'model_O4.onnx' | f013e34 |
| 2024-10-11 | Add exported ONNX model 'model_O3.onnx' | 91a47dc |
| 2024-10-11 | Add exported ONNX model 'model_O2.onnx' | fdd7846 |
| 2024-10-11 | Add exported ONNX model 'model_O1.onnx' | 1fc3e92 |
| 2024-10-11 | Add new SentenceTransformer model with an onnx backend | e15fb66 |
| 2024-03-27 | Adding `safetensors` variant of this model (#7) | 10b5b81 |
| 2024-02-12 | Update Sentence Transformers metadata (#5) | f2a4cb8 |
| 2022-06-15 | Add TF weights | 200b64f |
| 2021-11-02 | Update README.md | 9d5da8d |
| 2021-11-02 | Update README.md | 584f97e |
| 2021-11-02 | Update README.md | fc29aef |
| 2021-08-05 | Update README.md | 3f99938 |
| 2021-06-22 | Add new SentenceTransformer model. | d27719b |
| 2021-06-22 | initial commit | 40c3e43 |
All providers
2 providers · 2 endpoints
Where to run 2 providers · 2 endpoints
Hosted endpoints observed for clip-ViT-B-32-multilingual-v1, 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/clip-ViT-B-32-multilingual-v1 | 512 | — | $0.0050 | — | — | — | — | live | 2026-10-24 |
| HF Inference hf | sentence-transformers/clip-ViT-B-32-multilingual-v1 | — | — | — | — | — | — | — | 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
github.comauthority38 words
promptlayer.com255 words
clip-ViT-B-32-multilingual-v1 is a 135M parameter model released under the Apache 2.0 license that maps text and images into a shared vector space across over 50 languages. The implementation employs a DistilBERT foundation with a 128 token maximum sequence length and produces 512-dimensional output embeddings through multilingual knowledge distillation. This approach maintains the original CLIP image encoder while extending text capabilities for multilingual image search and zero-shot classification.written, reviewed, cited
deepinfra.com230 words
This model is a multilingual version of the OpenAI CLIP-ViT-B32 model that maps text and images to a common dense vector space. The text embedding model works for 50+ languages and was trained using Multilingual Knowledge Distillation with a multilingual DistilBERT student model.generated, cited
Papers
2 papers
Papers 2
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).
- Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation this model peer-reviewed · measuredNils Reimers, Iryna Gurevych · 2020-04-21 · Conference on Empirical Methods in Natural Language Processing (EMNLP) · PDF · HF paper page (1 upvotes) · evidence
Similar models
3 models
Similar models
- nomic-embed-text-v1.5 · 137M same size band (S), same task (sentence-similarity), permissive license
- distiluse-base-multilingual-cased-v2 · 135M same size band (S), same task (sentence-similarity), fewer providers, permissive license
- distiluse-base-multilingual-cased-v1 · 135M same size band (S), same task (sentence-similarity), fewer providers, permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:36 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:04 UTC · Last verified 2026-10-25
Follow clip-ViT-B-32-multilingual-v1
Watch this model
- /feeds/models/sentence-transformers--clip-ViT-B-32-multilingual-v1.xml(RSS 2.0) · every change to clip-ViT-B-32-multilingual-v1
- /feeds/families/sentence-transformers--clip-ViT.xml(RSS 2.0) · every change across the clip-ViT family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.