Scitopicnomicembed

🚀 nomic-ai/nomic-embed-text-v1.5

這是一個基於 sentence-transformers 的模型，它在 sci_topic_triplets 數據集上對 nomic-ai/nomic-embed-text-v1.5 進行了微調。該模型可以將句子和段落映射到一個 768 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

此模型基於 sentence-transformers 庫，在 sci_topic_triplets 數據集上對 nomic-ai/nomic-embed-text-v1.5 進行微調。它能將句子和段落映射到 768 維的密集向量空間，可用於語義文本相似度、語義搜索、釋義挖掘、文本分類、聚類等任務。

直接使用（Sentence Transformers）

首先安裝 Sentence Transformers 庫：

pip install -U sentence-transformers

然後，你可以加載這個模型並進行推理：

from sentence_transformers import SentenceTransformer

# 從 🤗 Hub 下載
model = SentenceTransformer("Corran/SciTopicNomicEmbed")
# 運行推理
sentences = [
    'The IANA Task Force (2021) builds upon previous research suggesting that slower gait speed is associated with increased risk of adverse outcomes in older adults (Levine et al., 2015; Schoenfeld et al., 2016).',
    'Gait speed at usual pace as a predictor of adverse outcomes in community-dwelling older people an International Academy on Nutrition and Aging (IANA) Task Force',
    'Referring to another writer’s idea(s) or position',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# 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

• 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。
• 高維向量映射：能夠將句子和段落映射到 768 維的密集向量空間，有效捕捉語義信息。
• 長序列處理：最大支持 8192 個標記的序列長度。

📦 安裝指南

首先安裝 Sentence Transformers 庫：

pip install -U sentence-transformers

💻 使用示例

基礎用法

from sentence_transformers import SentenceTransformer

# 從 🤗 Hub 下載
model = SentenceTransformer("Corran/SciTopicNomicEmbed")
# 運行推理
sentences = [
    'The IANA Task Force (2021) builds upon previous research suggesting that slower gait speed is associated with increased risk of adverse outcomes in older adults (Levine et al., 2015; Schoenfeld et al., 2016).',
    'Gait speed at usual pace as a predictor of adverse outcomes in community-dwelling older people an International Academy on Nutrition and Aging (IANA) Task Force',
    'Referring to another writer’s idea(s) or position',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 768]

# 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

模型來源

• 文檔：Sentence Transformers 文檔
• 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': 8192, 'do_lower_case': False}) with Transformer model: NomicBertModel 
  (1): Pooling({'word_embedding_dimension': 768, '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})
)

評估

指標

信息檢索

• 數據集：SciGen-Eval-Set
• 使用 InformationRetrievalEvaluator 進行評估

訓練詳情

訓練數據集

• sci_topic_triplets
  • 數據集：sci_topic_triplets（版本 8bf9936）
  • 大小：35,964 個訓練樣本
  • 列：query、positive 和 negative
  • 基於前 1000 個樣本的近似統計信息： | | 查詢 | 正樣本 | 負樣本 | | ---- | ---- | ---- | ---- | | 類型 | 字符串 | 字符串 | 字符串 | | 詳情 |
    • 最小：17 個標記
    • 平均：40.37 個標記
    • 最大：93 個標記
    |
    • 最小：5 個標記
    • 平均：18.75 個標記
    • 最大：56 個標記
    |
    • 最小：5 個標記
    • 平均：10.74 個標記
    • 最大：23 個標記
    |
  • 樣本： | 查詢 | 正樣本 | 負樣本 | | ---- | ---- | ---- | | This study provides comprehensive estimates of life expectancy, all-cause mortality, and cause-specific mortality for 249 causes of death and 195 countries and territories from 1980 to 2015, allowing for a detailed understanding of global health trends and patterns over the past four decades. | Global, regional, and national life expectancy, all-cause mortality, and cause-specific mortality for 249 causes of death, 1980–2015: a systematic analysis for the Global Burden of Disease Study 2015 | Explaining the significance of the current study | | This paper explores the relationship between the expected value and the volatility of the nominal excess return on stocks using a econometric approach. | On the Relation between the Expected Value and the Volatility of the Nominal Excess Return on Stocks | Stating the focus, aim, or argument of a short paper | | Despite the increasing attention given to the role of audit committees and board of directors in mitigating earnings management, several studies have reported inconclusive or even negative findings. | Audit committee, board of director characteristics, and earnings management | General reference to previous research or scholarship: highlighting negative outcomes |
  • 損失函數：MatryoshkaLoss，參數如下：

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768,
        384,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

評估數據集

• sci_topic_triplets
  • 數據集：sci_topic_triplets（版本 8bf9936）
  • 大小：4,495 個評估樣本
  • 列：query、positive 和 negative
  • 基於前 1000 個樣本的近似統計信息： | | 查詢 | 正樣本 | 負樣本 | | ---- | ---- | ---- | ---- | | 類型 | 字符串 | 字符串 | 字符串 | | 詳情 |
    • 最小：18 個標記
    • 平均：40.1 個標記
    • 最大：87 個標記
    |
    • 最小：5 個標記
    • 平均：18.75 個標記
    • 最大：58 個標記
    |
    • 最小：5 個標記
    • 平均：10.74 個標記
    • 最大：23 個標記
    |
  • 樣本： | 查詢 | 正樣本 | 負樣本 | | ---- | ---- | ---- | | In this cluster-randomised controlled trial, the authors aimed to evaluate the effectiveness of introducing the Medical Emergency Team (MET) system in reducing response times and improving patient outcomes in emergency departments. | Introduction of the medical emergency team (MET) system: a cluster-randomised controlled trial | Some ways of introducing quotations | | In the data collection phase of our study, we employed both surveys and interviews as research methods. Specifically, we administered surveys to 200 participants and conducted interviews with 10 key industry experts to gather proportional data on various aspects of management science practices. | Research Methodology: A Step-by-Step Guide for Beginners | Surveys and interviews: Reporting proportions | | Several density functional theory (DFT) based chemical reactivity indexes, such as the Fukui functions and the electrophilic and nucleophilic indices, are discussed in detail for their ability to predict chemical reactivity. | Chemical reactivity indexes in density functional theory | General comments on the relevant literature |
  • 損失函數：MatryoshkaLoss，參數如下：

{
    "loss": "MultipleNegativesRankingLoss",
    "matryoshka_dims": [
        768,
        384,
        256,
        128,
        64
    ],
    "matryoshka_weights": [
        1,
        1,
        1,
        1,
        1
    ],
    "n_dims_per_step": -1
}

訓練超參數

非默認超參數

• eval_strategy：按步驟評估
• per_device_train_batch_size：256
• per_device_eval_batch_size：256
• learning_rate：2e-05
• num_train_epochs：10
• warmup_ratio：0.1
• fp16：True
• load_best_model_at_end：True

所有超參數

訓練日誌

框架版本

• Python：3.11.11
• Sentence Transformers：3.3.1
• Transformers：4.47.1
• PyTorch：2.5.1+cu124
• Accelerate：1.2.1
• Datasets：3.2.0
• Tokenizers：0.21.0

📄 許可證

本模型使用 apache-2.0 許可證。

📖 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}
}