mmarco-mMiniLMv2-L12-H384-v1: Open-Weight Model

Runs On

What it takes to serve mmarco-mMiniLMv2-L12-H384-v1 (118M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

Precision	Weights	Memory needed	Cheapest setup	Per hour	Also fits
16-bit	0.2 GB	0.3 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
8-bit	0.1 GB	0.1 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00
4-bit	0.1 GB	0.1 GB	1x MI300X (192 GB) Vultr	$1.85	1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.

SAVRN's Notes on mmarco-mMiniLMv2-L12-H384-v1

Put this one behind the search index, not in front of it. It takes a query and candidate passages and ranks them, and because MS MARCO was machine translated into 14 languages for training, the same 118M parameters cover multilingual retrieval. At 16-bit the weights are 0.2 GB and it runs in 0.3 GB, so a 192 GB MI300X at $1.85 an hour is absurd on its own; park it beside whatever embedding or language model already holds the card.

Apache 2.0 covers commercial use, modification and redistribution if the license and NOTICE file travel with it and you state your changes. Two checks: the 514 token context must hold query and passage together, so chunk documents to fit, and the 4.1 GB download spans 23 files in safetensors, ONNX, OpenVINO and PyTorch, so take only the format your runtime loads. The base is mMiniLMv2-L12-H384, trained on unicamp-dl/mmarco.

SAVRN Research, 2026-09-18

Model Card

By Sentence Transformers - Cross-Encoders, published under apache-2.0, revision 1427fd652930.

Cross-Encoder for multilingual MS Marco

This model was trained on the MMARCO dataset. It is a machine translated version of MS MARCO using Google Translate. It was translated to 14 languages. In our experiments, we observed that it performs also well for other languages.

As a base model, we used the multilingual MiniLMv2 model.

The model can be used for Information Retrieval: Given a query, encode the query will all possible passages (e.g. retrieved with ElasticSearch). Then sort the passages in a decreasing order. See SBERT.net Retrieve & Re-rank for more details. The training code is available here: SBERT.net Training MS Marco

Usage with SentenceTransformers

The usage becomes easy when you have SentenceTransformers installed. Then, you can use the pre-trained models like this:

from sentence_transformers import CrossEncoder
model = CrossEncoder('model_name')
scores = model.predict([('Query', 'Paragraph1'), ('Query', 'Paragraph2') , ('Query', 'Paragraph3')])

Usage with Transformers

Read the full model card (205 words)

Configuration

Architecture: XLMRobertaForSequenceClassification
Context length (tokens): 514
Layers: 12
Hidden size: 384
Feed-forward size: 1,536
Attention heads: 12
Vocabulary size: 250,002
Stored precision: float32
Model type: xlm-roberta

Identity and Version

Repository: cross-encoder/mmarco-mMiniLMv2-L12-H384-v1
Publisher: Sentence Transformers - Cross-Encoders
Task: Text ranking
Modality: Other
Library: sentence-transformers
Parameters: 118M parameters
Languages: en, ar, zh, nl, fr, de, hi, in
Revision: 1427fd652930e4ba29e8149678df786c240d8825
First published: 2022-06-01
Last updated: 2025-04-15

Files and Weights

23 files, 4.1 GB in total. The weights are 13 files totalling 4.1 GB in bin, onnx, safetensors.

Weights13 files · 4.1 GB

Configuration3 files · 13.4 KB

Tokenizer2 files · 17.1 MB

Documentation1 file · 2.3 KB

Other3 files · 6.2 MB

Repository1 file · 1.2 KB

Every file

File	Type	Size	SHA-256
model.safetensors	Weights	470.6 MB	5daeca2481a7
onnx/model.onnx	Weights	470.9 MB	3e9a03ed1e96
onnx/model_O1.onnx	Weights	470.8 MB	6230c9f55c71
onnx/model_O2.onnx	Weights	470.6 MB	04aca53ef8e0
onnx/model_O3.onnx	Weights	470.6 MB	e3a4da3a7b0c
onnx/model_O4.onnx	Weights	235.5 MB	613b96ad598e
onnx/model_qint8_arm64.onnx	Weights	118.6 MB	1825907d6c1a
onnx/model_qint8_avx512.onnx	Weights	118.6 MB	1825907d6c1a
onnx/model_qint8_avx512_vnni.onnx	Weights	118.6 MB	1825907d6c1a
onnx/model_quint8_avx2.onnx	Weights	118.6 MB	6c2513767fb6
openvino/openvino_model.bin	Weights	470.6 MB	3b00b3db7a1a
openvino/openvino_model_qint8_quantized.bin	Weights	119.1 MB	2911002d538a
pytorch_model.bin	Weights	470.6 MB	1abc209e54d7
config.json	Configuration	891 B	—
special_tokens_map.json	Configuration	239 B	—
train_script.py	Configuration	12.2 KB	—
README.md	Documentation	2.3 KB	—
openvino/openvino_model.xml	Other	390.1 KB	—
openvino/openvino_model_qint8_quantized.xml	Other	711.8 KB	—
sentencepiece.bpe.model	Other	5.1 MB	cfc8146abe2a
.gitattributes	Repository	1.2 KB	—
tokenizer.json	Tokenizer	17.1 MB	62c24cdc13d4
tokenizer_config.json	Tokenizer	435 B	—

License and Download

License: apache-2.0
Access: Open weights, no gate
Download size: 4.1 GB

Download from Sentence Transformers - Cross-Encoders

Released by Sentence Transformers - Cross-Encoders through its official repository on Hugging Face. Read the license.

Built From

Derived from nreimers/mMiniLMv2-L12-H384-distilled-from-XLMR-Large
Quantized from nreimers/mMiniLMv2-L12-H384-distilled-from-XLMR-Large
Trained on (disclosed) unicamp-dl/mmarco

Memory Requirements

Precision	Weights in memory
As published	4.1 GB
16-bit	0.2 GB
8-bit	0.1 GB
4-bit	0.1 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Compare mmarco-mMiniLMv2-L12-H384-v1

Questions About mmarco-mMiniLMv2-L12-H384-v1

How much GPU memory does mmarco-mMiniLMv2-L12-H384-v1 need?

About 0.3 GB at 16-bit and 0.1 GB at 4-bit: the weights (118M parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run mmarco-mMiniLMv2-L12-H384-v1 on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use mmarco-mMiniLMv2-L12-H384-v1 commercially?

Yes. mmarco-mMiniLMv2-L12-H384-v1 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

What is mmarco-mMiniLMv2-L12-H384-v1's context length?

514 tokens, from the maximum position embeddings in its published configuration.

Similar Models

Model · Text ranking

ms-marco-electra-base

Sentence Transformers - Cross-Encoders

This model was trained on the MS Marco Passage Ranking task. The model can be used for Information Retrieval: Given a query, encode the query will all possible passages (e.g. retrieved with ElasticSearch). Then sort the passages in a decreasing order. See SBERT.net Retrieve & Re-rank for more details. The training code is available here: SBERT.net Training MS Marco The usage is easy when you have SentenceTransformers installed. Then you can use the pre-trained models like this: In the following table, we provide various pre-trained Cross-Encoders together with their performance on the TREC Deep Learning 2019 and the MS Marco Passage Reranking dataset.

Open weights apache-2.0 109M parameters 512 tokens sentence-transformers

View model

Model · Text ranking

qnli-electra-base

Sentence Transformers - Cross-Encoders

This model was trained using SentenceTransformers Cross-Encoder class. Given a question and paragraph, can the question be answered by the paragraph? The models have been trained on the GLUE QNLI dataset, which transformed the SQuAD dataset into an NLI task. For performance results of this model, see [SBERT.net Pre-trained Cross-Encoder][https://www.sbert.net/docs/pretrainedcross-encoders.html]. Pre-trained models can be used like this: You can use the model also directly with Transformers library (without SentenceTransformers library)

Open weights apache-2.0 109M parameters 512 tokens sentence-transformers

View model

Model · Text ranking

gte-reranker-modernbert-base

Alibaba-NLP

We are excited to introduce the gte-modernbert series of models, which are built upon the latest modernBERT pre-trained encoder-only foundation models. The gte-modernbert series models include both text embedding models and rerank models. The gte-modernbert models demonstrates competitive performance in several text embedding and text retrieval evaluation tasks when compared to similar-scale models from the current open-source community. This includes assessments such as MTEB, LoCO, and COIR evaluation. Use with transformers Use with sentence-transformers: Before you start, install the sentence-transformers libraries: Use with transformers.js Additionally, you can also deploy…

Open weights apache-2.0 150M parameters 8,192 tokens transformers

View model

Model · Text ranking

mxbai-rerank-xsmall-v1

Mixedbread

This is the smallest model in our family of powerful reranker models. You can learn more about the models in our blog post. Currently, the best way to use our models is with the most recent version of sentence-transformers. pip install -U sentence-transformers Let's say you have a query, and you want to rerank a set of documents. You can do that with only one line of code: Install transformers.js npm i @xenova/transformers Let's say you have a query, and you want to rerank a set of documents. In JavaScript, you need to add a function: You can use the large model via our API as follows: The API comes with additional features, such as a continous trained reranker! Check out the docs for more…

Open weights apache-2.0 71M parameters 512 tokens transformers

View model

Model · Text ranking

cross-encoder-russian-msmarco

Dmitry Tishencko

This is a sentence-transformers model based on a pre-trained DeepPavlov/rubert-base-cased and finetuned with MS-MARCO Russian passage ranking dataset. The model can be used for Information Retrieval in the Russian language: Given a query, encode the query will all possible passages (e.g. retrieved with ElasticSearch). Then sort the passages in a decreasing order. See SBERT.net Retrieve & Re-rank for more details. Using this model becomes easy when you have sentence-transformers installed: Then you can use the model like this: Without sentence-transformers, you can use the model like this: First, you pass your input through the transformer model, then you need to get the logits from the…

Open weights mit 178M parameters 512 tokens sentence-transformers

View model

Model · Text ranking

ms-marco-MiniLM-L12-v2

Sentence Transformers - Cross-Encoders

This model was trained on the MS Marco Passage Ranking task. The model can be used for Information Retrieval: Given a query, encode the query will all possible passages (e.g. retrieved with ElasticSearch). Then sort the passages in a decreasing order. See SBERT.net Retrieve & Re-rank for more details. The training code is available here: SBERT.net Training MS Marco The usage is easy when you have SentenceTransformers installed. Then you can use the pre-trained models like this: In the following table, we provide various pre-trained Cross-Encoders together with their performance on the TREC Deep Learning 2019 and the MS Marco Passage Reranking dataset.

Open weights apache-2.0 33M parameters 512 tokens sentence-transformers

View model