Finance2 Embedding Small En V1.5

🚀 基于BAAI/bge-small-en-v1.5的句子转换器

本模型是基于 Sentence Transformers 框架，在 baconnier/finance_dataset_small_private 数据集上对 BAAI/bge-small-en-v1.5 进行微调得到的。它能够将句子和段落映射到384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。

✨ 主要特性

• 基于预训练模型 BAAI/bge-small-en-v1.5 进行微调，在金融领域数据集上训练，能更好地处理金融相关文本。
• 可将文本映射到384维的密集向量空间，便于进行语义相似度计算等任务。
• 支持多种相似度计算函数，如余弦相似度。

📦 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 SentenceTransformer

# 从 🤗 Hub 下载模型
model = SentenceTransformer("baconnier/Finance2_embedding_small_en-V1.5")
# 进行推理
sentences = [
    'What is industrial production, and how is it measured by the Federal Reserve Board?',
    'Industrial production is a statistic determined by the Federal Reserve Board that measures the total output of all US factories and mines on a monthly basis. The Fed collects data from various government agencies and trade associations to calculate the industrial production index, which serves as an important economic indicator, providing insight into the health of the manufacturing and mining sectors.\nIndustrial production is a monthly statistic calculated by the Federal Reserve Board, measuring the total output of US factories and mines using data from government agencies and trade associations, serving as a key economic indicator for the manufacturing and mining sectors.',
    'Industrial production is a statistic that measures the output of factories and mines in the US. It is released by the Federal Reserve Board every quarter.\nIndustrial production measures factory and mine output, released quarterly by the Fed.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 384]

# 获取嵌入向量的相似度分数
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 文档
• 仓库：GitHub 上的 Sentence Transformers
• Hugging Face：Hugging Face 上的 Sentence Transformers

完整模型架构

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': True}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 384, '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

三元组

• 数据集：Finance_Embedding_Metric
• 使用 TripletEvaluator 进行评估

训练详情

训练数据集

• 数据集：baconnier/finance_dataset_small_private，版本 d7e6492
• 大小：15,525 个训练样本
• 列：anchor、positive 和 negative
• 损失函数：MultipleNegativesRankingLoss，参数如下：

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

评估数据集

• 数据集：baconnier/finance_dataset_small_private，版本 d7e6492
• 大小：862 个评估样本
• 列：anchor、positive 和 negative
• 损失函数：MultipleNegativesRankingLoss，参数如下：

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

训练超参数

非默认超参数

• eval_strategy: steps
• per_device_train_batch_size: 16
• per_device_eval_batch_size: 16
• num_train_epochs: 1
• warmup_ratio: 0.1
• bf16: True
• batch_sampler: no_duplicates

训练日志

框架版本

• Python: 3.10.12
• Sentence Transformers: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.3.0+cu121
• Accelerate: 0.31.0
• Datasets: 2.19.2
• Tokenizers: 0.19.1

📄 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",
}

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}
}