Fine Tuned Embedding Model

🚀 基於sentence-transformers/all-MiniLM-L6-v2的句子轉換器模型

本項目是一個基於 sentence-transformers 框架，從 sentence-transformers/all-MiniLM-L6-v2 微調而來的模型。它能將句子和段落映射到384維的密集向量空間，可用於語義文本相似度計算、語義搜索、釋義挖掘、文本分類、聚類等任務。

🚀 快速開始

本模型是基於 sentence-transformers 框架微調的，以下是使用該模型的具體步驟：

安裝Sentence Transformers庫

pip install -U sentence-transformers

加載模型並進行推理

from sentence_transformers import SentenceTransformer

# 從🤗 Hub下載模型
model = SentenceTransformer("sentence_transformers_model_id")
# 進行推理
sentences = [
    'What does this text say about data privacy?',
    'information during GAI training and maintenance. \nHuman-AI Configuration; Obscene, \nDegrading, and/or Abusive \nContent; Value Chain and \nComponent Integration; \nDangerous, Violent, or Hateful \nContent \nMS-2.6-002 \nAssess existence or levels of harmful bias, intellectual property infringement, \ndata privacy violations, obscenity, extremism, violence, or CBRN information in \nsystem training data. \nData Privacy; Intellectual Property; \nObscene, Degrading, and/or \nAbusive Content; Harmful Bias and \nHomogenization; Dangerous, \nViolent, or Hateful Content; CBRN \nInformation or Capabilities \nMS-2.6-003 Re-evaluate safety features of fine-tuned models when the negative risk exceeds \norganizational risk tolerance. \nDangerous, Violent, or Hateful \nContent \nMS-2.6-004 Review GAI system outputs for validity and safety: Review generated code to \nassess risks that may arise from unreliable downstream decision-making. \nValue Chain and Component \nIntegration; Dangerous, Violent, or \nHateful Content',
    'Scheurer, J. et al. (2023) Technical report: Large language models can strategically deceive their users \nwhen put under pressure. arXiv. https://arxiv.org/abs/2311.07590 \nShelby, R. et al. (2023) Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm \nReduction. arXiv. https://arxiv.org/pdf/2210.05791 \nShevlane, T. et al. (2023) Model evaluation for extreme risks. arXiv. https://arxiv.org/pdf/2305.15324 \nShumailov, I. et al. (2023) The curse of recursion: training on generated data makes models forget. arXiv. \nhttps://arxiv.org/pdf/2305.17493v2 \nSmith, A. et al. (2023) Hallucination or Confabulation? Neuroanatomy as metaphor in Large Language \nModels. PLOS Digital Health. \nhttps://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000388 \nSoice, E. et al. (2023) Can large language models democratize access to dual-use biotechnology? arXiv. \nhttps://arxiv.org/abs/2306.03809',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# 獲取嵌入向量的相似度分數
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

✨ 主要特性

• 高維向量映射：能夠將句子和段落映射到384維的密集向量空間。
• 多任務支持：可用於語義文本相似度計算、語義搜索、釋義挖掘、文本分類、聚類等多種自然語言處理任務。

📦 安裝指南

要使用本模型，需要安裝 sentence-transformers 庫，可使用以下命令進行安裝：

pip install -U sentence-transformers

💻 使用示例

基礎用法

from sentence_transformers import SentenceTransformer

# 從🤗 Hub下載模型
model = SentenceTransformer("sentence_transformers_model_id")
# 進行推理
sentences = [
    'What does this text say about data privacy?',
    'information during GAI training and maintenance. \nHuman-AI Configuration; Obscene, \nDegrading, and/or Abusive \nContent; Value Chain and \nComponent Integration; \nDangerous, Violent, or Hateful \nContent \nMS-2.6-002 \nAssess existence or levels of harmful bias, intellectual property infringement, \ndata privacy violations, obscenity, extremism, violence, or CBRN information in \nsystem training data. \nData Privacy; Intellectual Property; \nObscene, Degrading, and/or \nAbusive Content; Harmful Bias and \nHomogenization; Dangerous, \nViolent, or Hateful Content; CBRN \nInformation or Capabilities \nMS-2.6-003 Re-evaluate safety features of fine-tuned models when the negative risk exceeds \norganizational risk tolerance. \nDangerous, Violent, or Hateful \nContent \nMS-2.6-004 Review GAI system outputs for validity and safety: Review generated code to \nassess risks that may arise from unreliable downstream decision-making. \nValue Chain and Component \nIntegration; Dangerous, Violent, or \nHateful Content',
    'Scheurer, J. et al. (2023) Technical report: Large language models can strategically deceive their users \nwhen put under pressure. arXiv. https://arxiv.org/abs/2311.07590 \nShelby, R. et al. (2023) Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm \nReduction. arXiv. https://arxiv.org/pdf/2210.05791 \nShevlane, T. et al. (2023) Model evaluation for extreme risks. arXiv. https://arxiv.org/pdf/2305.15324 \nShumailov, I. et al. (2023) The curse of recursion: training on generated data makes models forget. arXiv. \nhttps://arxiv.org/pdf/2305.17493v2 \nSmith, A. et al. (2023) Hallucination or Confabulation? Neuroanatomy as metaphor in Large Language \nModels. PLOS Digital Health. \nhttps://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000388 \nSoice, E. et al. (2023) Can large language models democratize access to dual-use biotechnology? arXiv. \nhttps://arxiv.org/abs/2306.03809',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# 獲取嵌入向量的相似度分數
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]

📚 詳細文檔

模型詳情

模型描述

模型資源

• 文檔：Sentence Transformers文檔
• 代碼倉庫：GitHub上的Sentence Transformers
• Hugging Face：Hugging Face上的Sentence Transformers

完整模型架構

SentenceTransformer(
  (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, '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()
)

訓練詳情

訓練數據集

未命名數據集

• 數據集大小：555個訓練樣本
• 列信息：<code>sentence_0</code> 和 <code>sentence_1</code>
• 近似統計信息（基於前555個樣本）：

  	sentence_0	sentence_1
  類型	字符串	字符串
  詳情	最小：10個詞元 平均：11.2個詞元 最大：12個詞元	最小：156個詞元 平均：199.37個詞元 最大：256個詞元

• 樣本示例：

  sentence_0	sentence_1
  What does this text say about trustworthiness?	other systems. Information Integrity; Value Chain and Component Integration MP-2.2-002 Observe and analyze how the GAI system interacts with external networks, and identify any potential for negative externalities, particularly where content provenance might be compromised. Information Integrity AI Actor Tasks: End Users MAP 2.3: Scientific integrity and TEVV considerations are identified and documented, including those related to experimental design, data collection and selection (e.g., availability, representativeness, suitability), system trustworthiness, and construct validation Action ID Suggested Action GAI Risks MP-2.3-001 Assess the accuracy, quality, reliability, and authenticity of GAI output by comparing it to a set of known ground truth data and by using a variety of evaluation methods (e.g., human oversight and automated evaluation, proven cryptographic techniques, review of content inputs). Information Integrity 25
  What does this text say about unclassified?	training and TEVV data; Filtering of hate speech or content in GAI system training data; Prevalence of GAI-generated data in GAI system training data. Harmful Bias and Homogenization 15 Winogender Schemas is a sample set of paired sentences which differ only by gender of the pronouns used, which can be used to evaluate gender bias in natural language processing coreference resolution systems. 37 MS-2.11-005 Assess the proportion of synthetic to non-synthetic training data and verify training data is not overly homogenous or GAI-produced to mitigate concerns of model collapse. Harmful Bias and Homogenization AI Actor Tasks: AI Deployment, AI Impact Assessment, Affected Individuals and Communities, Domain Experts, End-Users, Operation and Monitoring, TEVV MEASURE 2.12: Environmental impact and sustainability of AI model training and management activities – as identified in the MAP function – are assessed and documented. Action ID Suggested Action GAI Risks
  What does this text say about unclassified?	Padmakumar, V. et al. (2024) Does writing with language models reduce content diversity? ICLR. https://arxiv.org/pdf/2309.05196 Park, P. et. al. (2024) AI deception: A survey of examples, risks, and potential solutions. Patterns, 5(5). arXiv. https://arxiv.org/pdf/2308.14752 Partnership on AI (2023) Building a Glossary for Synthetic Media Transparency Methods, Part 1: Indirect Disclosure. https://partnershiponai.org/glossary-for-synthetic-media-transparency-methods-part-1- indirect-disclosure/ Qu, Y. et al. (2023) Unsafe Diffusion: On the Generation of Unsafe Images and Hateful Memes From Text- To-Image Models. arXiv. https://arxiv.org/pdf/2305.13873 Rafat, K. et al. (2023) Mitigating carbon footprint for knowledge distillation based deep learning model compression. PLOS One. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0285668 Said, I. et al. (2022) Nonconsensual Distribution of Intimate Images: Exploring the Role of Legal Attitudes

訓練損失函數

使用 MultipleNegativesRankingLoss 損失函數，參數如下：

{
    "scale": 20.0,
    "similarity_fct": "cos_sim"
}

訓練超參數

非默認超參數

• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• multi_dataset_batch_sampler: round_robin

所有超參數

框架版本

• Python：3.11.5
• Sentence Transformers：3.1.1
• Transformers：4.44.2
• PyTorch：2.4.1+cpu
• Accelerate：0.34.2
• Datasets：3.0.0
• Tokenizers：0.19.1

🔧 技術細節

本模型基於 sentence-transformers 框架，從 sentence-transformers/all-MiniLM-L6-v2 微調而來。模型結構包含 Transformer 層、Pooling 層和 Normalize 層，具體如下：

SentenceTransformer(
  (0): Transformer({'max_seq_length': 256, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, '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()
)

在訓練過程中，使用了 MultipleNegativesRankingLoss 損失函數，並設置了相應的超參數，以優化模型性能。

📄 許可證

文檔中未提及相關許可證信息。

📖 引用

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",
}

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}
}