Sbert Pq

🚀 SBERT_PQ

SBERT_PQ is a sentence-transformers model designed to determine the relevance between a short text (mostly a single sentence with 10 - 15 words) and a question.

The model computes 312-dimensional vectors for both the text and the question. The cosine of the angle between these vectors provides an assessment of whether the text contains an answer to the given question. In the dialogue system project, it is used for semantic search of fact records in the database based on the question posed by the interlocutor.

🚀 Quick Start

✨ Features

• Based on a Small Model: The model is built on cointegrated/rubert-tiny2, featuring a very small size and fast inference even on CPU.
• High Accuracy: The maximum value of the cossim_f1 metric on the test set (10% of the dataset) is 0.986. When using the sberbank-ai/ruBert-base model as the base, the maximum cossim_f1 is 0.992.

📦 Installation

You need to install sentence-transformers:

pip install -U sentence-transformers

💻 Usage Examples

Basic Usage

import sentence_transformers

sentences = ["Кошка ловит мышку.", "Чем занята кошка?"]

model = sentence_transformers.SentenceTransformer('inkoziev/sbert_pq')
embeddings = model.encode(sentences)

s = sentence_transformers.util.cos_sim(a=embeddings[0], b=embeddings[1])
print('text={} question={} cossim={}'.format(sentences[0], sentences[1], s))

📄 License

This project is under the unlicense license.

📚 Documentation

Citation

@MISC{rugpt_chitchat,
    author  = {Ilya Koziev},
    title   = {Texts & Questions Relevancy Model},
    url     = {https://huggingface.co/inkoziev/sbert_pq},
    year    = 2022
}

Information Table