PyLate model based on nomic-ai/nomic-embed-text-v2-moe
This is a PyLate model finetuned from nomic-ai/nomic-embed-text-v2-moe. It maps sentences & paragraphs to sequences of 128-dimensional dense vectors and can be used for semantic textual similarity using the MaxSim operator.
PyLate provides a streamlined interface to index and retrieve documents using ColBERT models. The index leverages the Voyager HNSW index to efficiently handle document embeddings and enable fast retrieval.
Indexing documents
First, load the ColBERT model and initialize the Voyager index, then encode and index your documents:
python
1from pylate import indexes, models, retrieve
23# Step 1: Load the ColBERT model4model = models.ColBERT(5 model_name_or_path=pylate_model_id,6)78# Step 2: Initialize the Voyager index9index = indexes.Voyager(10 index_folder="pylate-index",11 index_name="index",12 override=True,# This overwrites the existing index if any13)1415# Step 3: Encode the documents16documents_ids =["1","2","3"]17documents =["document 1 text","document 2 text","document 3 text"]1819documents_embeddings = model.encode(20 documents,21 batch_size=32,22 is_query=False,# Ensure that it is set to False to indicate that these are documents, not queries23 show_progress_bar=True,24)2526# Step 4: Add document embeddings to the index by providing embeddings and corresponding ids27index.add_documents(28 documents_ids=documents_ids,29 documents_embeddings=documents_embeddings,30)
Note that you do not have to recreate the index and encode the documents every time. Once you have created an index and added the documents, you can re-use the index later by loading it:
python
1# To load an index, simply instantiate it with the correct folder/name and without overriding it2index = indexes.Voyager(3 index_folder="pylate-index",4 index_name="index",5)
Retrieving top-k documents for queries
Once the documents are indexed, you can retrieve the top-k most relevant documents for a given set of queries.
To do so, initialize the ColBERT retriever with the index you want to search in, encode the queries and then retrieve the top-k documents to get the top matches ids and relevance scores:
python
1# Step 1: Initialize the ColBERT retriever2retriever = retrieve.ColBERT(index=index)34# Step 2: Encode the queries5queries_embeddings = model.encode(6["query for document 3","query for document 1"],7 batch_size=32,8 is_query=True,# # Ensure that it is set to False to indicate that these are queries9 show_progress_bar=True,10)1112# Step 3: Retrieve top-k documents13scores = retriever.retrieve(14 queries_embeddings=queries_embeddings,15 k=10,# Retrieve the top 10 matches for each query16)
Reranking
If you only want to use the ColBERT model to perform reranking on top of your first-stage retrieval pipeline without building an index, you can simply use rank function and pass the queries and documents to rerank:
Approximate statistics based on the first 1000 samples:
query
positive
negative
type
string
string
string
details
min: 6 tokens
mean: 22.92 tokens
max: 32 tokens
min: 5 tokens
mean: 16.66 tokens
max: 32 tokens
min: 5 tokens
mean: 15.57 tokens
max: 32 tokens
Samples:
query
positive
negative
هل رأيت الآنسة (ديزي) تقود
هل حصلت على مشاهدة القيادة الآنسة ديزي.
أنا سعيد لأننا شاهدنا "قيادة الآنسة (ديزي) " سوياً
ونعم يا (ستيف) ، أريد أن أسمع نظريتك السياسية لـ (فيل هاريس).
(ستيف) لديه نظرية (فيل هاريس) للسياسة
الأخ والأخت يتعلمون القراءة
هكذا احتفل آل توكوجاوا بدين أسلافهم الشينتو المتمجد بالمزارات الفخمة التي بنوها في نيكو
دين الشنتو كان يحتفل به توكوجاوا
التوكوغاوا لم يبنوا أي معابد شنتو
Loss: pylate.losses.contrastive.Contrastive
Evaluation Dataset
Unnamed Dataset
Size: 4,149 evaluation samples
Columns: query, positive, and negative
Approximate statistics based on the first 1000 samples:
query
positive
negative
type
string
string
string
details
min: 6 tokens
mean: 23.0 tokens
max: 32 tokens
min: 5 tokens
mean: 16.79 tokens
max: 32 tokens
min: 4 tokens
mean: 15.35 tokens
max: 32 tokens
Samples:
query
positive
negative
كيوتو هي المركز الوطني لتلك التخصصات التقليدية مثل تشا دو (احتفال الشاي) وإيكيبانا (ترتيب الزهور) ، وموطن ولادة الكابوكي، والمركز الرائد للكتابة الخطية والرسم والنحت.
يتم ممارسة الأنشطة التقليدية ، مثل ترتيب الزهور وحفلات الشاي ، في كيوتو.
(راؤول) قام بخيانة الساقي ولم يعط الساقي بقشيشاً واحداً
انت تعلم انهم مازالوا مدمنين للمخدرات ولكنهم شرعيين
إنها عادة مخدرات قانونية لكنها لا تزال عادة مخدرات
امرأة تقف على شاطئ رملي
زورق نهر أزرق مليء بالمرأة يطفو أمام زورق أصفر آخر
1@inproceedings{reimers-2019-sentence-bert,
2 title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
3 author = "Reimers, Nils and Gurevych, Iryna",
4 booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
5 month = "11",
6 year = "2019",
7 publisher = "Association for Computational Linguistics",
8 url = "https://arxiv.org/abs/1908.10084"
9}
PyLate
bibtex
1@misc{PyLate,
2title={PyLate: Flexible Training and Retrieval for Late Interaction Models},
3author={Chaffin, Antoine and Sourty, Raphaël},
4url={https://github.com/lightonai/pylate},
5year={2024}
6}
