Snoweu V2

🚀 基于Snowflake/snowflake-arctic-embed-m-v1.5的句子转换器

这是一个基于 sentence-transformers 框架，从 Snowflake/snowflake-arctic-embed-m-v1.5 微调而来的模型。它可以将句子和段落映射到一个768维的密集向量空间，可用于语义文本相似度计算、语义搜索、释义挖掘、文本分类、聚类等任务。

🚀 快速开始

安装依赖库

首先，你需要安装 sentence-transformers 库：

pip install -U sentence-transformers

加载模型并进行推理

from sentence_transformers import SentenceTransformer

# 从🤗 Hub下载模型
model = SentenceTransformer("sentence_transformers_model_id")
# 进行推理
sentences = [
    'What is the definition of a preliminary economic assessment in the context of evaluating projects for the recovery of critical raw materials?',
    '(39)\n\n‘preliminary economic assessment’ means an early-stage, conceptual assessment of the potential economic viability of a project for the recovery of critical raw materials from extractive waste;\n\n(40)\n\n‘magnetic resonance imaging device’ means a non-invasive medical device that uses magnetic fields to make anatomical images or any other device that uses magnetic fields to make images of the inside of object;\n\n(41)\n\n‘wind energy generator’ means the part of an onshore or offshore wind turbine that converts the mechanical energy of the rotor into electrical energy;\n\n(42)',
    'For the purposes of the first subparagraph of this paragraph, insurance undertakings referred to in point (a) of the first subparagraph of Article 1(3) of this Directive that are part of a group, on the basis of financial relationships referred to in point (c)(ii) of Article 212(1) of Directive 2009/138/EC, and which are subject to group supervision in accordance with points (a) to (c) of Article 213(2) of that Directive shall be treated as subsidiary undertakings of the parent undertaking of that group.\n\n9.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# 获取嵌入向量的相似度得分
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

✨ 主要特性

• 多任务支持：可用于语义文本相似度计算、语义搜索、释义挖掘、文本分类、聚类等多种自然语言处理任务。
• 高维向量表示：将句子和段落映射到768维的密集向量空间，便于进行语义分析。
• 微调模型：基于 Snowflake/snowflake-arctic-embed-m-v1.5 进行微调，能更好地适应特定任务。

📦 安装指南

安装 sentence-transformers 库：

pip install -U sentence-transformers

💻 使用示例

基础用法

from sentence_transformers import SentenceTransformer

# 从🤗 Hub下载模型
model = SentenceTransformer("sentence_transformers_model_id")
# 进行推理
sentences = [
    'What is the definition of a preliminary economic assessment in the context of evaluating projects for the recovery of critical raw materials?',
    '(39)\n\n‘preliminary economic assessment’ means an early-stage, conceptual assessment of the potential economic viability of a project for the recovery of critical raw materials from extractive waste;\n\n(40)\n\n‘magnetic resonance imaging device’ means a non-invasive medical device that uses magnetic fields to make anatomical images or any other device that uses magnetic fields to make images of the inside of object;\n\n(41)\n\n‘wind energy generator’ means the part of an onshore or offshore wind turbine that converts the mechanical energy of the rotor into electrical energy;\n\n(42)',
    'For the purposes of the first subparagraph of this paragraph, insurance undertakings referred to in point (a) of the first subparagraph of Article 1(3) of this Directive that are part of a group, on the basis of financial relationships referred to in point (c)(ii) of Article 212(1) of Directive 2009/138/EC, and which are subject to group supervision in accordance with points (a) to (c) of Article 213(2) of that Directive shall be treated as subsidiary undertakings of the parent undertaking of that group.\n\n9.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# 获取嵌入向量的相似度得分
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

📚 详细文档

模型详情

模型描述

模型来源

• 文档：Sentence Transformers Documentation
• 仓库：Sentence Transformers on GitHub
• Hugging Face：Sentence Transformers on Hugging Face

完整模型架构

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': True, 'pooling_mode_mean_tokens': False, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
  (2): Normalize()
)

评估

指标

信息检索

使用 InformationRetrievalEvaluator 进行评估。

训练详情

训练数据集

未命名数据集

• 大小：29,911个训练样本
• 列：sentence_0 和 sentence_1
• 近似统计信息（基于前1000个样本）：

  	sentence_0	sentence_1
  类型	字符串	字符串
  详情	最小值：13个标记 平均值：41.63个标记 最大值：252个标记	最小值：4个标记 平均值：233.72个标记 最大值：512个标记

• 样本：

  sentence_0	sentence_1
  What measures must Member States take to ensure that workers who believe they have been discriminated against in terms of equal pay can establish their case before a competent authority or national court?	Article 18 Shift of burden of proof 1. Member States shall take the appropriate measures, in accordance with their national judicial systems, to ensure that, when workers who consider themselves wronged because the principle of equal pay has not been applied to them establish before a competent authority or national court facts from which it may be presumed that there has been direct or indirect discrimination, it shall be for the respondent to prove that there has been no direct or indirect discrimination in relation to pay. 2. Member States shall ensure that, in administrative procedures or court proceedings regarding alleged direct or indirect discrimination in relation to pay, where an employer has not implemented the pay transparency obligations set out in Articles 5, 6, 7, 9 and 10, it is for the employer to prove that there has been no such discrimination. The first subparagraph of this paragraph shall not apply where the employer proves that the infringement of the obligati...
  What are the key considerations for recognizing and addressing discrimination in the context of compensation and penalties, particularly in relation to the gender pay gap?	discrimination, in particular for substantive and procedural purposes, including to recognise the existence of discrimination, to decide on the appropriate comparator, to assess the proportionality, and to determine, where relevant, the level of compensation awarded or penalties imposed. An intersectional approach is important for understanding and addressing the gender pay gap. This clarification should not change the scope of employers’ obligations in regard to the pay transparency measures under this Directive. In particular, employers should not be required to gather data related to protected grounds other than sex.
  What is the process for aircraft operators and shipping companies regarding the surrendering of allowances in relation to their total emissions from the previous calendar year?	(b) each aircraft operator surrenders a number of allowances that is equal to its total emissions during the preceding calendar year, as verified in accordance with Article 15; (c) each shipping company surrenders a number of allowances that is equal to its total emissions during the preceding calendar year, as verified in accordance with Article 3ge. Member States, administering Member States and administering authorities in respect of a shipping company shall ensure that allowances surrendered in accordance with the first subparagraph are subsequently cancelled. ▼M15 3-e.

训练超参数

非默认超参数

• eval_strategy: steps
• per_device_train_batch_size: 6
• per_device_eval_batch_size: 6
• num_train_epochs: 4
• multi_dataset_batch_sampler: round_robin

框架版本

• Python：3.10.11
• Sentence Transformers：3.4.1
• Transformers：4.48.1
• PyTorch：2.4.0+cu121
• Accelerate：1.4.0
• Datasets：3.3.2
• Tokenizers：0.21.0

🔧 技术细节

损失函数

使用 MatryoshkaLoss 损失函数，参数如下：

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768,
        512,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

训练日志

📄 许可证

文档中未提及相关许可证信息。

📖 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",
}

MatryoshkaLoss

@misc{kusupati2024matryoshka,
    title={Matryoshka Representation Learning},
    author={Aditya Kusupati and Gantavya Bhatt and Aniket Rege and Matthew Wallingford and Aditya Sinha and Vivek Ramanujan and William Howard-Snyder and Kaifeng Chen and Sham Kakade and Prateek Jain and Ali Farhadi},
    year={2024},
    eprint={2205.13147},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

MultipleNegativesRankingLoss

@misc{henderson2017efficient,
    title={Efficient Natural Language Response Suggestion for Smart Reply},
    author={Matthew Henderson and Rami Al-Rfou and Brian Strope and Yun-hsuan Sung and Laszlo Lukacs and Ruiqi Guo and Sanjiv Kumar and Balint Miklos and Ray Kurzweil},
    year={2017},
    eprint={1705.00652},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}