Reranker Gte Multilingual Base Msmarco Bce Ep 2
基于sentence-transformers库在msmarco数据集上训练的交叉编码器模型,用于文本重排序和语义搜索
Downloads 28
Release Time : 4/6/2025
Model Overview
该模型计算文本对的分数,可用于文本重排序和语义搜索任务。它是在199万条训练样本上使用二元交叉熵损失函数训练得到的。
Model Features
高性能重排序
在NanoMSMARCO_R100数据集上达到0.7008的NDCG@10分数,表现优异
大规模训练
使用199万条训练样本进行训练,具有强大的语义理解能力
长文本支持
最大支持8192个标记的序列长度,适合处理长文本
Model Capabilities
文本对评分
语义搜索
搜索结果重排序
Use Cases
信息检索
搜索引擎结果重排序
对搜索引擎返回的结果进行重新排序,提升相关性
在MSMARCO数据集上NDCG@10达到0.7008
问答系统
答案相关性排序
对候选答案进行相关性排序,选择最佳答案
在NanoNQ_R100数据集上NDCG@10达到0.6888
🚀 CrossEncoder
CrossEncoder 是一个基于 sentence-transformers 库,在 msmarco 数据集上训练的 Cross Encoder 模型。它可以计算文本对的得分,可用于文本重排序和语义搜索。
✨ 主要特性
- 能够计算文本对的得分,适用于文本重排序和语义搜索任务。
- 支持长序列输入,最大序列长度可达 8192 个标记。
📦 安装指南
首先,安装 Sentence Transformers 库:
pip install -U sentence-transformers
💻 使用示例
基础用法
from sentence_transformers import CrossEncoder
# 从 🤗 Hub 下载模型
model = CrossEncoder("skfrost19/reranker-gte-multilingual-base-msmarco-bce-ep-2")
# 获取文本对的得分
pairs = [
['what symptoms might a patient with a tmd have', 'TMD sufferers have a long list of symptoms, including chronic pain (https://youtu.be/SvMaJb8o2RI), many of which are in common with Parkinsonâ\x80\x99s disease (PD) symptoms.'],
['what is a thermal protector', 'The word hero comes from the Greek á¼¥Ï\x81Ï\x89Ï\x82 (hÄ\x93rÅ\x8ds), hero, warrior, particularly one such as Heracles with divine ancestry or later given divine honors. literally protector or defender.'],
['how many copies of call of duty wwii sold', 'Call of Duty 3. Call of Duty 3 is a World War II first-person shooter and the third installment in the Call of Duty video game series. Released on November 7, 2006, the game was developed by Treyarch, and was the first major installment in the Call of Duty series not to be developed by Infinity Ward. It was also the first not to be released on the PC platform. It was released on the PlayStation 2, PlayStation 3, Wii, Xbox, and Xbox 360.'],
['what is the desired temperature for the fresh food compartment in a refrigerator', 'A refrigerator maintains a temperature a few degrees above the freezing point of water. Optimum temperature range for perishable food storage is 3 to 5 °C (37 to 41 °F).emperature settings for refrigerator and freezer compartments are often given arbitrary numbers by manufacturers (for example, 1 through 9, warmest to coldest), but generally 3 to 5 °C (37 to 41 °F) is ideal for the refrigerator compartment and â\x88\x9218 °C (0 °F) for the freezer.'],
['what is gsm alarm system', 'Iâ\x80\x99m sure you would have these questions in your mind when you heard GSM alarm system at the first time. GSM alarm system is an alarm system that operating through GSM (global system for mobile communications) network; not requiring a telephone line.urthermore, in the case of burglar entering the premises and cutting the telephone line, the GSM alarm would not be affected and still work as it does not require the use of a fixed phone line. So this security alarm is ideal for the place where no fixed phone line or hard to get one.'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)
# 或者根据与单个文本的相似度对不同文本进行排序
ranks = model.rank(
'what symptoms might a patient with a tmd have',
[
'TMD sufferers have a long list of symptoms, including chronic pain (https://youtu.be/SvMaJb8o2RI), many of which are in common with Parkinsonâ\x80\x99s disease (PD) symptoms.',
'The word hero comes from the Greek á¼¥Ï\x81Ï\x89Ï\x82 (hÄ\x93rÅ\x8ds), hero, warrior, particularly one such as Heracles with divine ancestry or later given divine honors. literally protector or defender.',
'Call of Duty 3. Call of Duty 3 is a World War II first-person shooter and the third installment in the Call of Duty video game series. Released on November 7, 2006, the game was developed by Treyarch, and was the first major installment in the Call of Duty series not to be developed by Infinity Ward. It was also the first not to be released on the PC platform. It was released on the PlayStation 2, PlayStation 3, Wii, Xbox, and Xbox 360.',
'A refrigerator maintains a temperature a few degrees above the freezing point of water. Optimum temperature range for perishable food storage is 3 to 5 °C (37 to 41 °F).emperature settings for refrigerator and freezer compartments are often given arbitrary numbers by manufacturers (for example, 1 through 9, warmest to coldest), but generally 3 to 5 °C (37 to 41 °F) is ideal for the refrigerator compartment and â\x88\x9218 °C (0 °F) for the freezer.',
'Iâ\x80\x99m sure you would have these questions in your mind when you heard GSM alarm system at the first time. GSM alarm system is an alarm system that operating through GSM (global system for mobile communications) network; not requiring a telephone line.urthermore, in the case of burglar entering the premises and cutting the telephone line, the GSM alarm would not be affected and still work as it does not require the use of a fixed phone line. So this security alarm is ideal for the place where no fixed phone line or hard to get one.',
]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]
📚 详细文档
模型详情
模型描述
属性 | 详情 |
---|---|
模型类型 | Cross Encoder |
最大序列长度 | 8192 个标记 |
输出标签数量 | 1 个标签 |
训练数据集 | msmarco |
语言 | 英语 |
模型来源
- 文档:Sentence Transformers 文档
- 文档:Cross Encoder 文档
- 仓库:GitHub 上的 Sentence Transformers
- Hugging Face:Hugging Face 上的 Cross Encoders
评估
评估指标
Cross Encoder 重排序
- 数据集:
NanoMSMARCO_R100
、NanoNFCorpus_R100
和NanoNQ_R100
- 评估方法:使用
CrossEncoderRerankingEvaluator
进行评估,参数如下:
{
"at_k": 10,
"always_rerank_positives": true
}
指标 | NanoMSMARCO_R100 | NanoNFCorpus_R100 | NanoNQ_R100 |
---|---|---|---|
map | 0.6439 (+0.1543) | 0.3324 (+0.0714) | 0.6174 (+0.1978) |
mrr@10 | 0.6382 (+0.1607) | 0.5372 (+0.0374) | 0.6412 (+0.2145) |
ndcg@10 | 0.7008 (+0.1604) | 0.3732 (+0.0482) | 0.6888 (+0.1881) |
Cross Encoder Nano BEIR
- 数据集:
NanoBEIR_R100_mean
- 评估方法:使用
CrossEncoderNanoBEIREvaluator
进行评估,参数如下:
{
"dataset_names": [
"msmarco",
"nfcorpus",
"nq"
],
"rerank_k": 100,
"at_k": 10,
"always_rerank_positives": true
}
指标 | 值 |
---|---|
map | 0.5312 (+0.1412) |
mrr@10 | 0.6055 (+0.1375) |
ndcg@10 | 0.5876 (+0.1322) |
训练详情
训练数据集
- 数据集:msmarco,版本为 9e329ed
- 大小:1,990,000 个训练样本
- 列:
query
、passage
和score
- 近似统计信息(基于前 1000 个样本):
| | 查询 | 段落 | 得分 |
| ---- | ---- | ---- | ---- |
| 类型 | 字符串 | 字符串 | 浮点数 |
| 详情 |
- 最小长度:11 个字符
- 平均长度:34.61 个字符
- 最大长度:124 个字符
- 最小长度:82 个字符
- 平均长度:357.43 个字符
- 最大长度:1034 个字符
- 最小值:0.0
- 平均值:0.49
- 最大值:1.0
- 样本:
| 查询 | 段落 | 得分 |
| ---- | ---- | ---- |
|
what causes your tailbone to hurt
|A coccyx injury results in pain and discomfort in the tailbone area (the condition is called coccydynia). These injuries may result in a bruise, dislocation, or fracture (break) of the coccyx. Although they may be slow to heal, the majority of coccyx injuries can be managed with cautious treatment.ost tailbone injuries are caused by trauma to the coccyx area. 1 A fall onto the tailbone in the seated position, usually against a hard surface, is the most common cause of coccyx injuries. 2 A direct blow to the tailbone, such as those that occur during contact sports, can injure the coccyx.
|1.0
| |what muscles do trunk lateral flexion
|Itâs the same with the External Obliques, but unlike the External Obliques, they are not visible when fully developed. Action: 1 Supports abdominal wall, assists forced respiration, aids raising intra-abdominal pressure and, with muscles of other side, abducts and rotates trunk. 2 Contraction of one side alone laterally bends the trunk to that side and rotates the trunk to the other side.
|0.0
| |brake horsepower definition
|When the brake lights will not come on, the first thing to check is the third-brake light. If it too is not working, the brake-light switch, a bad fuse or an unplugged harness is likely.ull up on the brake pedal and if the lights go out, switch mis-alignment or pedal position error is the likely cause. The final possibility is a wire shorted to power. Unplug the brake-light switch and if the lights stay on, a short circuit is the case.
|0.0
| - 损失函数:
BinaryCrossEntropyLoss
,参数如下:
{
"activation_fn": "torch.nn.modules.linear.Identity",
"pos_weight": null
}
评估数据集
- 数据集:msmarco,版本为 9e329ed
- 大小:10,000 个评估样本
- 列:
query
、passage
和score
- 近似统计信息(基于前 1000 个样本):
| | 查询 | 段落 | 得分 |
| ---- | ---- | ---- | ---- |
| 类型 | 字符串 | 字符串 | 浮点数 |
| 详情 |
- 最小长度:9 个字符
- 平均长度:33.72 个字符
- 最大长度:193 个字符
- 最小长度:55 个字符
- 平均长度:353.35 个字符
- 最大长度:895 个字符
- 最小值:0.0
- 平均值:0.5
- 最大值:1.0
- 样本:
| 查询 | 段落 | 得分 |
| ---- | ---- | ---- |
|
what symptoms might a patient with a tmd have
|TMD sufferers have a long list of symptoms, including chronic pain (https://youtu.be/SvMaJb8o2RI), many of which are in common with Parkinsonâs disease (PD) symptoms.
|1.0
| |what is a thermal protector
|The word hero comes from the Greek á¼¥ÏÏÏ (hÄrÅs), hero, warrior, particularly one such as Heracles with divine ancestry or later given divine honors. literally protector or defender.
|0.0
| |how many copies of call of duty wwii sold
|Call of Duty 3. Call of Duty 3 is a World War II first-person shooter and the third installment in the Call of Duty video game series. Released on November 7, 2006, the game was developed by Treyarch, and was the first major installment in the Call of Duty series not to be developed by Infinity Ward. It was also the first not to be released on the PC platform. It was released on the PlayStation 2, PlayStation 3, Wii, Xbox, and Xbox 360.
|0.0
| - 损失函数:
BinaryCrossEntropyLoss
,参数如下:
{
"activation_fn": "torch.nn.modules.linear.Identity",
"pos_weight": null
}
训练超参数
非默认超参数
eval_strategy
: stepsper_device_train_batch_size
: 156per_device_eval_batch_size
: 156learning_rate
: 2e-05num_train_epochs
: 2warmup_ratio
: 0.1seed
: 12bf16
: Truedataloader_num_workers
: 4load_best_model_at_end
: True
所有超参数
点击展开
overwrite_output_dir
: Falsedo_predict
: Falseeval_strategy
: stepsprediction_loss_only
: Trueper_device_train_batch_size
: 156per_device_eval_batch_size
: 156per_gpu_train_batch_size
: Noneper_gpu_eval_batch_size
: Nonegradient_accumulation_steps
: 1eval_accumulation_steps
: Nonetorch_empty_cache_steps
: Nonelearning_rate
: 2e-05weight_decay
: 0.0adam_beta1
: 0.9adam_beta2
: 0.999adam_epsilon
: 1e-08max_grad_norm
: 1.0num_train_epochs
: 2max_steps
: -1lr_scheduler_type
: linearlr_scheduler_kwargs
: {}warmup_ratio
: 0.1warmup_steps
: 0log_level
: passivelog_level_replica
: warninglog_on_each_node
: Truelogging_nan_inf_filter
: Truesave_safetensors
: Truesave_on_each_node
: Falsesave_only_model
: Falserestore_callback_states_from_checkpoint
: Falseno_cuda
: Falseuse_cpu
: Falseuse_mps_device
: Falseseed
: 12data_seed
: Nonejit_mode_eval
: Falseuse_ipex
: Falsebf16
: Truefp16
: Falsefp16_opt_level
: O1half_precision_backend
: autobf16_full_eval
: Falsefp16_full_eval
: Falsetf32
: Nonelocal_rank
: 0ddp_backend
: Nonetpu_num_cores
: Nonetpu_metrics_debug
: Falsedebug
: []dataloader_drop_last
: Falsedataloader_num_workers
: 4dataloader_prefetch_factor
: Nonepast_index
: -1disable_tqdm
: Falseremove_unused_columns
: Truelabel_names
: Noneload_best_model_at_end
: Trueignore_data_skip
: Falsefsdp
: []fsdp_min_num_params
: 0fsdp_config
: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}tp_size
: 0fsdp_transformer_layer_cls_to_wrap
: Noneaccelerator_config
: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}deepspeed
: Nonelabel_smoothing_factor
: 0.0optim
: adamw_torchoptim_args
: Noneadafactor
: Falsegroup_by_length
: Falselength_column_name
: lengthddp_find_unused_parameters
: Noneddp_bucket_cap_mb
: Noneddp_broadcast_buffers
: Falsedataloader_pin_memory
: Truedataloader_persistent_workers
: Falseskip_memory_metrics
: Trueuse_legacy_prediction_loop
: Falsepush_to_hub
: Falseresume_from_checkpoint
: Nonehub_model_id
: Nonehub_strategy
: every_savehub_private_repo
: Nonehub_always_push
: Falsegradient_checkpointing
: Falsegradient_checkpointing_kwargs
: Noneinclude_inputs_for_metrics
: Falseinclude_for_metrics
: []eval_do_concat_batches
: Truefp16_backend
: autopush_to_hub_model_id
: Nonepush_to_hub_organization
: Nonemp_parameters
:auto_find_batch_size
: Falsefull_determinism
: Falsetorchdynamo
: Noneray_scope
: lastddp_timeout
: 1800torch_compile
: Falsetorch_compile_backend
: Nonetorch_compile_mode
: Nonedispatch_batches
: Nonesplit_batches
: Noneinclude_tokens_per_second
: Falseinclude_num_input_tokens_seen
: Falseneftune_noise_alpha
: Noneoptim_target_modules
: Nonebatch_eval_metrics
: Falseeval_on_start
: Falseuse_liger_kernel
: Falseeval_use_gather_object
: Falseaverage_tokens_across_devices
: Falseprompts
: Nonebatch_sampler
: batch_samplermulti_dataset_batch_sampler
: proportional
训练日志
轮次 | 步数 | 训练损失 | 验证损失 | NanoMSMARCO_R100_ndcg@10 | NanoNFCorpus_R100_ndcg@10 | NanoNQ_R100_ndcg@10 | NanoBEIR_R100_mean_ndcg@10 |
---|---|---|---|---|---|---|---|
-1 | -1 | - | - | 0.6561 (+0.1157) | 0.3777 (+0.0527) | 0.6548 (+0.1541) | 0.5629 (+0.1075) |
0.0001 | 1 | 0.1089 | - | - | - | - | - |
0.3136 | 4000 | 0.1258 | - | - | - | - | - |
0.6271 | 8000 | 0.1185 | - | - | - | - | - |
0.7839 | 10000 | - | 0.1177 | 0.7008 (+0.1604) | 0.3732 (+0.0482) | 0.6888 (+0.1881) | 0.5876 (+0.1322) |
0.9407 | 12000 | 0.1231 | - | - | - | - | - |
-1 | -1 | - | - | 0.7008 (+0.1604) | 0.3732 (+0.0482) | 0.6888 (+0.1881) | 0.5876 (+0.1322) |
注:加粗行表示保存的检查点。
框架版本
- Python: 3.11.5
- Sentence Transformers: 4.0.1
- Transformers: 4.50.3
- PyTorch: 2.6.0+cu124
- Accelerate: 1.6.0
- Datasets: 3.5.0
- Tokenizers: 0.21.1
📄 许可证
文档中未提及该模型的许可证信息。
📚 引用
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",
}
Jina Embeddings V3
Jina Embeddings V3 是一个多语言句子嵌入模型,支持超过100种语言,专注于句子相似度和特征提取任务。
文本嵌入
Transformers Supports Multiple Languages

J
jinaai
3.7M
911
Ms Marco MiniLM L6 V2
Apache-2.0
基于MS Marco段落排序任务训练的交叉编码器模型,用于信息检索中的查询-段落相关性评分
文本嵌入 English
M
cross-encoder
2.5M
86
Opensearch Neural Sparse Encoding Doc V2 Distill
Apache-2.0
基于蒸馏技术的稀疏检索模型,专为OpenSearch优化,支持免推理文档编码,在搜索相关性和效率上优于V1版本
文本嵌入
Transformers English

O
opensearch-project
1.8M
7
Sapbert From PubMedBERT Fulltext
Apache-2.0
基于PubMedBERT的生物医学实体表征模型,通过自对齐预训练优化语义关系捕捉
文本嵌入 English
S
cambridgeltl
1.7M
49
Gte Large
MIT
GTE-Large 是一个强大的句子转换器模型,专注于句子相似度和文本嵌入任务,在多个基准测试中表现出色。
文本嵌入 English
G
thenlper
1.5M
278
Gte Base En V1.5
Apache-2.0
GTE-base-en-v1.5 是一个英文句子转换器模型,专注于句子相似度任务,在多个文本嵌入基准测试中表现优异。
文本嵌入
Transformers Supports Multiple Languages

G
Alibaba-NLP
1.5M
63
Gte Multilingual Base
Apache-2.0
GTE Multilingual Base 是一个多语言的句子嵌入模型,支持超过50种语言,适用于句子相似度计算等任务。
文本嵌入
Transformers Supports Multiple Languages

G
Alibaba-NLP
1.2M
246
Polybert
polyBERT是一个化学语言模型,旨在实现完全由机器驱动的超快聚合物信息学。它将PSMILES字符串映射为600维密集指纹,以数值形式表示聚合物化学结构。
文本嵌入
Transformers

P
kuelumbus
1.0M
5
Bert Base Turkish Cased Mean Nli Stsb Tr
Apache-2.0
基于土耳其语BERT的句子嵌入模型,专为语义相似度任务优化
文本嵌入
Transformers Other

B
emrecan
1.0M
40
GIST Small Embedding V0
MIT
基于BAAI/bge-small-en-v1.5模型微调的文本嵌入模型,通过MEDI数据集与MTEB分类任务数据集训练,优化了检索任务的查询编码能力。
文本嵌入
Safetensors English
G
avsolatorio
945.68k
29
Featured Recommended AI Models
Llama 3 Typhoon V1.5x 8b Instruct
专为泰语设计的80亿参数指令模型,性能媲美GPT-3.5-turbo,优化了应用场景、检索增强生成、受限生成和推理任务
大型语言模型
Transformers Supports Multiple Languages

L
scb10x
3,269
16
Cadet Tiny
Openrail
Cadet-Tiny是一个基于SODA数据集训练的超小型对话模型,专为边缘设备推理设计,体积仅为Cosmo-3B模型的2%左右。
对话系统
Transformers English

C
ToddGoldfarb
2,691
6
Roberta Base Chinese Extractive Qa
基于RoBERTa架构的中文抽取式问答模型,适用于从给定文本中提取答案的任务。
问答系统 Chinese
R
uer
2,694
98