/ Модель

Характеристики

Контекст
512 токенов
Параметры
124 млн
Архитектура
bert · dense
Базовая модель
Multilingual-MiniLM-L12-H384
Библиотека
sentence-transformers
Опубликована
20 ноя 2025

reranker-Multilingual-MiniLM-L12-H384-msmarco-mse

Ранжирование текстаТекст → score8 профессий

Англоязычный кросс-энкодер для оценки релевантности пар «запрос — текст» и переранжирования поисковой выдачи.

29скачиваний за 30 дней0лайковвеса обновлены20 ноя 2025

Бенчмарки

Как модель справилась с задачами пользователей ailist и что о ней публикуют открытые лидерборды

Тесты пользователейпока нет

Никто ещё не опубликовал тест этой модели. Соберите бенчмарк из своих задач — результат появится здесь.

  1. Добавьте модель в бенчмарк

    Вместе с другими моделями категории «Ранжирование» — сравнение идёт внутри одной категории.

  2. Опишите свои задачи

    Напишите запросы, при необходимости приложите файлы и эталонные ответы.

  3. Запустите и опубликуйте

    Модели ответят на каждую задачу, судья оценит ответы. Опубликованный результат появится на этой странице.

Публичные оценки

Из открытых лидербордов. Сравнивайте результаты одного теста при одинаковых условиях — отсутствие оценки не значит, что модель хуже.

В подключённых источниках результатов для этой модели пока нет.

Что умеет модель

Описание составлено по карточке модели и метаданным Hugging Face

Модель hasankursun/reranker-Multilingual-MiniLM-L12-H384-msmarco-mse — кросс-энкодер, дообученный на MS MARCO на базе microsoft/Multilingual-MiniLM-L12-H384 с помощью библиотеки Sentence Transformers. Она оценивает релевантность пары «запрос — текст» одним числовым баллом и предназначена для переранжирования результатов поиска и семантического поиска, в том числе как второй этап после более быстрого первичного поиска. Модель содержит 124,3 млн параметров, принимает текстовые пары и имеет максимальную длину последовательности 512 токенов.

В README приведены результаты оценки на наборах NanoMSMARCO, NanoNFCorpus, NanoNQ и NanoBEIR; заявленный язык модели — английский. Выходной балл подходит для сортировки кандидатов по релевантности. Лицензия в предоставленных данных не указана.

Сценарии применения

  • переранжирование найденных документов по запросу
  • оценка релевантности отрывков для семантического поиска
  • сортировка ответов-кандидатов перед подачей в RAG-систему

Ограничения

Максимальная длина последовательности составляет 512 токенов; модель оценивает пары текстов и требует отдельного списка кандидатов для переранжирования. В README заявлен английский язык, поэтому качество для других языков не подтверждено.

Для каких профессий подходит8

Разделы каталога профессий, в задачах которых используется эта модель

Все профессии

Также подходит для

Карточка модели

README от автора на Hugging Face, как он его опубликовал

Открыть на Hugging Face

CrossEncoder based on microsoft/Multilingual-MiniLM-L12-H384

This is a Cross Encoder model finetuned from microsoft/Multilingual-MiniLM-L12-H384 on the msmarco dataset using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

Model Description

Model Sources

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("hasankursun/reranker-Multilingual-MiniLM-L12-H384-msmarco-mse")
# Get scores for pairs of texts
pairs = [
    ['what who did dr nassar molest', 'Dr. Larry Nassar, who is 54 years old, has been in prison while awaiting sentencing. During that time, McKayla Maroney and Aly Raisman revealed that Dr. Nassar had molested them, sharing their heartbreaking #MeToo stories with the world. Like so many others, these Olympic superstars describe Dr. Nassar, at the time their team doctor, administering treatments to them under the guise of medical practice.'],
    ['average monthly temperature in vancouver', 'Vancouver: Annual Weather Averages. August is the hottest month in Vancouver with an average temperature of 18°C (64°F) and the coldest is January at 4°C (38°F) with the most daily sunshine hours at 13 in July.The wettest month is November with an average of 200mm of rain.The best month to swim in the sea is in July when the average sea temperature is 14°C (57°F).he wettest month is November with an average of 200mm of rain. The best month to swim in the sea is in July when the average sea temperature is 14°C (57°F).'],
    ['do i have to file a tax return for my business if i dont have any income', 'The responsibility for filing your childâ\x80\x99s tax return rests with your child if he is capable of doing so. If he is not old enough to understand how to prepare a tax return, then it becomes your responsibility to file it for him or to include his income on your return.he responsibility for filing your childâ\x80\x99s tax return rests with your child if he is capable of doing so. If he is not old enough to understand how to prepare a tax return, then it becomes your responsibility to file it for him or to include his income on your return.'],
    ['what happens to your growth plate when you finish growing', 'Over the past 70 million years, the combined processes of magma formation, volcano eruption and growth, and continued movement of the Pacific Plate over the stationary Hawaiian hot-spot have left a long trail of volcanoes across the Pacific Ocean floor.s the Pacific Plate continues to move west-northwest, the Island of Hawaii will be carried beyond the hotspot by plate motion, setting the stage for the formation of a new volcanic island in its place.'],
    ['average cost of holiday to new zealand', 'Definition: Average total cost is the sum of all the production costs divided by the number of units produced. Terms related to Average Total Cost: 1  The Cost Curve. 2  Average Propensity to Consume.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    'what who did dr nassar molest',
    [
        'Dr. Larry Nassar, who is 54 years old, has been in prison while awaiting sentencing. During that time, McKayla Maroney and Aly Raisman revealed that Dr. Nassar had molested them, sharing their heartbreaking #MeToo stories with the world. Like so many others, these Olympic superstars describe Dr. Nassar, at the time their team doctor, administering treatments to them under the guise of medical practice.',
        'Vancouver: Annual Weather Averages. August is the hottest month in Vancouver with an average temperature of 18°C (64°F) and the coldest is January at 4°C (38°F) with the most daily sunshine hours at 13 in July.The wettest month is November with an average of 200mm of rain.The best month to swim in the sea is in July when the average sea temperature is 14°C (57°F).he wettest month is November with an average of 200mm of rain. The best month to swim in the sea is in July when the average sea temperature is 14°C (57°F).',
        'The responsibility for filing your childâ\x80\x99s tax return rests with your child if he is capable of doing so. If he is not old enough to understand how to prepare a tax return, then it becomes your responsibility to file it for him or to include his income on your return.he responsibility for filing your childâ\x80\x99s tax return rests with your child if he is capable of doing so. If he is not old enough to understand how to prepare a tax return, then it becomes your responsibility to file it for him or to include his income on your return.',
        'Over the past 70 million years, the combined processes of magma formation, volcano eruption and growth, and continued movement of the Pacific Plate over the stationary Hawaiian hot-spot have left a long trail of volcanoes across the Pacific Ocean floor.s the Pacific Plate continues to move west-northwest, the Island of Hawaii will be carried beyond the hotspot by plate motion, setting the stage for the formation of a new volcanic island in its place.',
        'Definition: Average total cost is the sum of all the production costs divided by the number of units produced. Terms related to Average Total Cost: 1  The Cost Curve. 2  Average Propensity to Consume.',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Evaluation

Metrics

Cross Encoder Reranking

  • Datasets: NanoMSMARCO_R100, NanoNFCorpus_R100 and NanoNQ_R100
  • Evaluated with CrossEncoderRerankingEvaluator with these parameters:
    {
        "at_k": 10,
        "always_rerank_positives": true
    }
    
MetricNanoMSMARCO_R100NanoNFCorpus_R100NanoNQ_R100
map0.5969 (+0.1073)0.3452 (+0.0842)0.6475 (+0.2279)
mrr@100.5916 (+0.1141)0.5892 (+0.0893)0.6618 (+0.2351)
ndcg@100.6640 (+0.1236)0.3925 (+0.0674)0.7035 (+0.2028)

Cross Encoder Nano BEIR

  • Dataset: NanoBEIR_R100_mean
  • Evaluated with CrossEncoderNanoBEIREvaluator with these parameters:
    {
        "dataset_names": [
            "msmarco",
            "nfcorpus",
            "nq"
        ],
        "rerank_k": 100,
        "at_k": 10,
        "always_rerank_positives": true
    }
    
MetricValue
map0.5299 (+0.1398)
mrr@100.6142 (+0.1462)
ndcg@100.5867 (+0.1313)

Training Details

Training Dataset

msmarco

  • Dataset: msmarco at 9e329ed
  • Size: 1,990,000 training samples
  • Columns: score, query, and passage
  • Approximate statistics based on the first 1000 samples:
    scorequerypassage
    typefloatstringstring
    detailsmin: -11.82mean: 0.84max: 10.86min: 10 charactersmean: 33.38 charactersmax: 91 charactersmin: 57 charactersmean: 346.19 charactersmax: 916 characters
  • Samples:
    scorequerypassage
    -1.2796382506688435what hormone is released after a mealThis conduction of bile is the main function of the common bile duct. The hormone cholecystokinin, when stimulated by a fatty meal, promotes bile secretion by increased production of hepatic bile, contraction of the gall bladder, and relaxation of the Sphincter of Oddi.
    5.993939399719238what can cause fluid around the heart and in the lungsBut in certain circumstances, the alveoli fill with fluid instead of air, preventing oxygen from being absorbed into your bloodstream. A number of things can cause fluid to accumulate in your lungs, but most have to do with your heart (cardiogenic pulmonary edema).
    -3.6289062102635703which war did augustus underwood serve in?Quick Answer. Caesar Augustus, also known as Octavian, was the first Roman emperor after the assassination of Julius Caesar in 43 B.C. Augustus was Caesar's grand nephew who shrewdly combined lawmaking, military might and institutional building to create the foundations of the 200-year Pax Romana, explains Biography.com.
  • Loss: MSELoss with these parameters:
    {
        "activation_fn": "torch.nn.modules.linear.Identity"
    }
    

Evaluation Dataset

msmarco

  • Dataset: msmarco at 9e329ed
  • Size: 10,000 evaluation samples
  • Columns: score, query, and passage
  • Approximate statistics based on the first 1000 samples:
    scorequerypassage
    typefloatstringstring
    detailsmin: -11.82mean: 0.75max: 11.1min: 10 charactersmean: 33.5 charactersmax: 173 charactersmin: 50 charactersmean: 350.19 charactersmax: 1175 characters
  • Samples:
    scorequerypassage
    7.531805515289307what who did dr nassar molestDr. Larry Nassar, who is 54 years old, has been in prison while awaiting sentencing. During that time, McKayla Maroney and Aly Raisman revealed that Dr. Nassar had molested them, sharing their heartbreaking #MeToo stories with the world. Like so many others, these Olympic superstars describe Dr. Nassar, at the time their team doctor, administering treatments to them under the guise of medical practice.
    8.721138636271158average monthly temperature in vancouverVancouver: Annual Weather Averages. August is the hottest month in Vancouver with an average temperature of 18°C (64°F) and the coldest is January at 4°C (38°F) with the most daily sunshine hours at 13 in July.The wettest month is November with an average of 200mm of rain.The best month to swim in the sea is in July when the average sea temperature is 14°C (57°F).he wettest month is November with an average of 200mm of rain. The best month to swim in the sea is in July when the average sea temperature is 14°C (57°F).
    -2.6914831002553306do i have to file a tax return for my business if i dont have any incomeThe responsibility for filing your child’s tax return rests with your child if he is capable of doing so. If he is not old enough to understand how to prepare a tax return, then it becomes your responsibility to file it for him or to include his income on your return.he responsibility for filing your child’s tax return rests with your child if he is capable of doing so. If he is not old enough to understand how to prepare a tax return, then it becomes your responsibility to file it for him or to include his income on your return.
  • Loss: MSELoss with these parameters:
    {
        "activation_fn": "torch.nn.modules.linear.Identity"
    }
    

Training Hyperparameters

Non-Default Hyperparameters

  • eval_strategy: steps
  • per_device_train_batch_size: 16
  • per_device_eval_batch_size: 16
  • learning_rate: 8e-06
  • num_train_epochs: 1
  • warmup_ratio: 0.1
  • seed: 12
  • bf16: True
  • dataloader_num_workers: 4
  • load_best_model_at_end: True

All Hyperparameters

  • overwrite_output_dir: False
  • do_predict: False
  • eval_strategy: steps
  • prediction_loss_only: True
  • per_device_train_batch_size: 16
  • per_device_eval_batch_size: 16
  • per_gpu_train_batch_size: None
  • per_gpu_eval_batch_size: None
  • gradient_accumulation_steps: 1
  • eval_accumulation_steps: None
  • torch_empty_cache_steps: None
  • learning_rate: 8e-06
  • weight_decay: 0.0
  • adam_beta1: 0.9
  • adam_beta2: 0.999
  • adam_epsilon: 1e-08
  • max_grad_norm: 1.0
  • num_train_epochs: 1
  • max_steps: -1
  • lr_scheduler_type: linear
  • lr_scheduler_kwargs: {}
  • warmup_ratio: 0.1
  • warmup_steps: 0
  • log_level: passive
  • log_level_replica: warning
  • log_on_each_node: True
  • logging_nan_inf_filter: True
  • save_safetensors: True
  • save_on_each_node: False
  • save_only_model: False
  • restore_callback_states_from_checkpoint: False
  • no_cuda: False
  • use_cpu: False
  • use_mps_device: False
  • seed: 12
  • data_seed: None
  • jit_mode_eval: False
  • bf16: True
  • fp16: False
  • fp16_opt_level: O1
  • half_precision_backend: auto
  • bf16_full_eval: False
  • fp16_full_eval: False
  • tf32: None
  • local_rank: 0
  • ddp_backend: None
  • tpu_num_cores: None
  • tpu_metrics_debug: False
  • debug: []
  • dataloader_drop_last: False
  • dataloader_num_workers: 4
  • dataloader_prefetch_factor: None
  • past_index: -1
  • disable_tqdm: False
  • remove_unused_columns: True
  • label_names: None
  • load_best_model_at_end: True
  • ignore_data_skip: False
  • fsdp: []
  • fsdp_min_num_params: 0
  • fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
  • fsdp_transformer_layer_cls_to_wrap: None
  • accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
  • parallelism_config: None
  • deepspeed: None
  • label_smoothing_factor: 0.0
  • optim: adamw_torch_fused
  • optim_args: None
  • adafactor: False
  • group_by_length: False
  • length_column_name: length
  • project: huggingface
  • trackio_space_id: trackio
  • ddp_find_unused_parameters: None
  • ddp_bucket_cap_mb: None
  • ddp_broadcast_buffers: False
  • dataloader_pin_memory: True
  • dataloader_persistent_workers: False
  • skip_memory_metrics: True
  • use_legacy_prediction_loop: False
  • push_to_hub: False
  • resume_from_checkpoint: None
  • hub_model_id: None
  • hub_strategy: every_save
  • hub_private_repo: None
  • hub_always_push: False
  • hub_revision: None
  • gradient_checkpointing: False
  • gradient_checkpointing_kwargs: None
  • include_inputs_for_metrics: False
  • include_for_metrics: []
  • eval_do_concat_batches: True
  • fp16_backend: auto
  • push_to_hub_model_id: None
  • push_to_hub_organization: None
  • mp_parameters:
  • auto_find_batch_size: False
  • full_determinism: False
  • torchdynamo: None
  • ray_scope: last
  • ddp_timeout: 1800
  • torch_compile: False
  • torch_compile_backend: None
  • torch_compile_mode: None
  • include_tokens_per_second: False
  • include_num_input_tokens_seen: no
  • neftune_noise_alpha: None
  • optim_target_modules: None
  • batch_eval_metrics: False
  • eval_on_start: False
  • use_liger_kernel: False
  • liger_kernel_config: None
  • eval_use_gather_object: False
  • average_tokens_across_devices: True
  • prompts: None
  • batch_sampler: batch_sampler
  • multi_dataset_batch_sampler: proportional
  • router_mapping: {}
  • learning_rate_mapping: {}

Training Logs

EpochStepTraining LossValidation LossNanoMSMARCO_R100_ndcg@10NanoNFCorpus_R100_ndcg@10NanoNQ_R100_ndcg@10NanoBEIR_R100_mean_ndcg@10
-1-1--0.0412 (-0.4992)0.2383 (-0.0867)0.0468 (-0.4538)0.1088 (-0.3466)
0.0000166.3043-----
0.0322400051.8043-----
0.0643800019.8202-----
0.0965120008.3776-----
0.1286160005.8256-----
0.1608200004.90584.17380.6666 (+0.1262)0.3713 (+0.0463)0.6888 (+0.1882)0.5756 (+0.1202)
0.1930240004.4046-----
0.2251280004.0191-----
0.2573320003.8165-----
0.2894360003.5786-----
0.3216400003.4843.34110.6856 (+0.1451)0.3848 (+0.0597)0.6858 (+0.1852)0.5854 (+0.1300)
0.3538440003.3467-----
0.3859480003.1952-----
0.4181520003.1518-----
0.4503560003.0006-----
0.4824600003.00282.93310.6609 (+0.1204)0.3894 (+0.0643)0.7033 (+0.2027)0.5845 (+0.1292)
0.5146640002.8912-----
0.5467680002.8112-----
0.5789720002.7707-----
0.6111760002.7566-----
0.6432800002.6672.62700.6603 (+0.1199)0.3967 (+0.0717)0.6790 (+0.1784)0.5787 (+0.1233)
0.6754840002.6895-----
0.7075880002.6151-----
0.7397920002.558-----
0.7719960002.5346-----
0.8041000002.52552.38770.6640 (+0.1236)0.3925 (+0.0674)0.7035 (+0.2028)0.5867 (+0.1313)
0.83621040002.488-----
0.86831080002.5009-----
0.90051120002.4195-----
0.93271160002.4237-----
0.96481200002.46712.36690.6640 (+0.1236)0.3955 (+0.0705)0.6877 (+0.1871)0.5824 (+0.1270)
0.99701240002.3769-----
-1-1--0.6640 (+0.1236)0.3925 (+0.0674)0.7035 (+0.2028)0.5867 (+0.1313)
  • The bold row denotes the saved checkpoint.

Framework Versions

  • Python: 3.12.3
  • Sentence Transformers: 5.1.2
  • Transformers: 4.57.1
  • PyTorch: 2.9.1+cu128
  • Accelerate: 1.11.0
  • Datasets: 4.4.1
  • Tokenizers: 0.22.1

Citation

BibTeX

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084",
}

Как запустить

На своей GPU в ailist. Цена — по текущим предложениям провайдеров

GPU в ailist

Универсальный рантайм · sentence-transformers

Можно запустить

Подбираем GPU…

Класс архитектуры
BertForSequenceClassification
Ревизия весов
4024278

0 моделей