Fingumv3

🚀 基於 dunzhang/stella_en_1.5B_v5 的句子轉換器

這是一個基於 dunzhang/stella_en_1.5B_v5 微調的 句子轉換器 模型。它可以將句子和段落映射到一個 1024 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

直接使用（句子轉換器）

首先安裝句子轉換器庫：

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("sentence_transformers_model_id")
# 運行推理
sentences = [
    'The Tchaikovsky Symphony Orchestra is a Russian classical music orchestra established in 1930. It was founded as the Moscow Radio Symphony Orchestra, and served as the official symphony for the Soviet All-Union Radio network. Following the dissolution of the, Soviet Union in 1991, the orchestra was renamed in 1993 by the Russian Ministry of Culture in recognition of the central role the music of Tchaikovsky plays in its repertoire. The current music director is Vladimir Fedoseyev, who has been in that position since 1974.',
    'Instruct: Given a web search query, retrieve relevant passages that answer the query.\nQuery: Tchaikovsky Symphony Orchestra',
    'Instruct: Given a web search query, retrieve relevant passages that answer the query.\nQuery: Sierra del Lacandón',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# 獲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

該模型具有以下特點：

• 基於 dunzhang/stella_en_1.5B_v5 進行微調。
• 能夠將句子和段落映射到 1024 維的密集向量空間。
• 可用於多種自然語言處理任務，如語義文本相似度計算、語義搜索等。

📦 安裝指南

若要使用該模型，需先安裝 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("sentence_transformers_model_id")
# 運行推理
sentences = [
    'The Tchaikovsky Symphony Orchestra is a Russian classical music orchestra established in 1930. It was founded as the Moscow Radio Symphony Orchestra, and served as the official symphony for the Soviet All-Union Radio network. Following the dissolution of the, Soviet Union in 1991, the orchestra was renamed in 1993 by the Russian Ministry of Culture in recognition of the central role the music of Tchaikovsky plays in its repertoire. The current music director is Vladimir Fedoseyev, who has been in that position since 1974.',
    'Instruct: Given a web search query, retrieve relevant passages that answer the query.\nQuery: Tchaikovsky Symphony Orchestra',
    'Instruct: Given a web search query, retrieve relevant passages that answer the query.\nQuery: Sierra del Lacandón',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 1024]

# 獲取嵌入向量的相似度得分
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：句子轉換器文檔
• 倉庫：GitHub 上的句子轉換器
• Hugging Face：Hugging Face 上的句子轉換器

完整模型架構

SentenceTransformer(
  (0): Transformer({'max_seq_length': 8096, 'do_lower_case': False}) with Transformer model: Qwen2Model 
  (1): Pooling({'word_embedding_dimension': 1536, '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): Dense({'in_features': 1536, 'out_features': 1024, 'bias': True, 'activation_function': 'torch.nn.modules.linear.Identity'})
)

評估

信息檢索

使用 InformationRetrievalEvaluator 進行評估。

訓練詳情

訓練超參數

非默認超參數

• eval_strategy: steps
• per_device_eval_batch_size: 4
• gradient_accumulation_steps: 4
• learning_rate: 2e-05
• max_steps: 1500
• lr_scheduler_type: cosine
• warmup_ratio: 0.1
• warmup_steps: 5
• bf16: True
• tf32: True
• optim: adamw_torch_fused
• gradient_checkpointing: True
• gradient_checkpointing_kwargs: {'use_reentrant': False}
• batch_sampler: no_duplicates

所有超參數

訓練日誌

框架版本

• Python: 3.10.12
• 句子轉換器: 3.0.1
• Transformers: 4.41.2
• PyTorch: 2.2.0+cu121
• Accelerate: 0.33.0
• Datasets: 2.20.0
• Tokenizers: 0.19.1

🔧 技術細節

該模型基於 dunzhang/stella_en_1.5B_v5 進行微調，使用了 SentenceTransformer 架構。在訓練過程中，採用了多種超參數進行優化，如學習率、批量大小等。同時，使用了 MatryoshkaLoss 和 MultipleNegativesRankingLoss 作為損失函數。

📄 許可證

文檔中未提及相關許可證信息。

📖 引用

BibTeX

句子轉換器

@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}
}