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; 1 quantized build for local use.
Generated 2026-10-07 15:03 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.
Weights-only memory for distiluse-base-multilingual-cased-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.
Hosted endpoints observed for distiluse-base-multilingual-cased-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.
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.
Sentence Transformers released distiluse-base-multilingual-cased-v1, an open-weight model for sentence similarity under the Apache License 2.0. The 135M parameter model features a 512-token context and maps text to a 512-dimensional vector space for clustering or semantic search. At 16-bit precision, it requires about 0.3 GB of GPU memory, fitting on a single MI300X rented for $1.85 an hour, and draws 1.5M downloads a month.written, reviewed, cited
The model requires chunking longer passages because its 512-token context limit is strict.
Sentence Transformers built the model on DistilBERT using a vocabulary with 119,547 entries.
The model ships in multiple formats including safetensors, ONNX, OpenVINO, PyTorch, and TensorFlow.
sentence-transformers/distiluse-base-multilingual-cased-v1 maps sentences to a 512 dimensional dense vector space for clustering or semantic search. The model architecture includes a Transformer with max_seq_length: 128 and do_lower_case: False, followed by mean token pooling and a Dense layer with Tanh activation. It has recorded 68,568 downloads last month.written, reviewed, cited
github.comauthorityabout the distiluse family676 words
The distiluse-base-multilingual-cased model adds a dense layer on top of mean pooling to transform DistilBERT's 768-dimensional output into 512 dimensions, matching the Universal Sentence Encoder's format. This architecture allows the model to function as a student in a knowledge distillation setup where the teacher is the multilingual Universal Sentence Encoder. The method enables extending sentence embedding models to new languages using parallel training data.written, reviewed, cited
The distiluse-base-multilingual-cased model uses the training process from the paper Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation.
github.comauthorityabout the distiluse family929 words
The distiluse-base-multilingual-cased model outputs a 512 dimension tensor instead of the typical 768 because it uses a dense layer to down-project the 768 dimension output to 512 dimensions. This down projection was necessary as the model is a distillation of the multilingual Universal Sentence Encoder which uses 512 dimension. The developer plans to add the models to the huggingface models collection within the next 2 weeks.written, reviewed, cited
github.com2020-10-18authorityabout the distiluse family696 words
distiluse-base-multilingual-cased supports 50+ languages out-of-the-box for tasks like STS or duplicate detection without further training. The model includes English plus 53 additional languages, though performance drops for those without parallel data. Training on new tasks is possible but not required for standard use cases.written, reviewed, cited
The model is much better for languages where you have parallel data compared to languages where you don't have parallel data.
The listed languages in the documentation are the additional languages besides English, meaning English is supported by default.
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).
