Vectorizer V1 S En

🚀 Model Card for vectorizer-v1-S-en

This model, developed by Sinequa, is a vectorizer that generates an embedding vector for a given passage or query. Passage vectors are stored in our vector index, and the query vector is used at query time to find relevant passages in the index.

🚀 Quick Start

The vectorizer-v1-S-en model is ready to generate embedding vectors for passages and queries right away. Just input your text, and it will output the corresponding vector.

✨ Features

• Multilingual Support: Trained and tested in English, making it suitable for English-language tasks.
• Efficient Inference: Offers fast inference times on various GPUs with different quantization types and batch sizes.
• Low Memory Usage: Consumes relatively low GPU memory, especially when using FP16 quantization.

📦 Installation

Requirements

• Minimal Sinequa version: 11.10.0
• Minimal Sinequa version for using FP16 models and GPUs with CUDA compute capability of 8.9+ (like NVIDIA L4): 11.11.0
• Cuda compute capability: above 5.0 (above 6.0 for FP16 use)

💻 Usage Examples

Basic Usage

# Assume there is a Python API to use the model
from sinequa_vectorizer import Vectorizer

vectorizer = Vectorizer(model_name='vectorizer-v1-S-en')
passage = "This is a sample passage."
passage_vector = vectorizer.get_vector(passage)
print(passage_vector)

📚 Documentation

Supported Languages

The model was trained and tested in the following languages:

• English

Scores

Note that the relevance score is computed as an average over 14 retrieval datasets (see details below).

Inference Times

GPU Memory usage

Note that GPU memory usage only includes how much GPU memory the actual model consumes on an NVIDIA T4 GPU with a batch size of 32. It does not include the fix amount of memory that is consumed by the ONNX Runtime upon initialization which can be around 0.5 to 1 GiB depending on the used GPU.

Model Details

Overview

• Number of parameters: 29 million
• Base language model: English BERT-Small
• Insensitive to casing and accents
• Output dimensions: 256 (reduced with an additional dense layer)
• Training procedure: A first model was trained with query-passage pairs, using the in-batch negative strategy with this loss. A second model was then trained on query-passage-negative triplets with negatives mined from the previous model, like a variant of ANCE but with different hyper parameters.

Training Data

The model was trained on a Sinequa curated version of Google's Natural Questions.

Evaluation Metrics

To determine the relevance score, we averaged the results that we obtained when evaluating on the datasets of the BEIR benchmark. Note that all these datasets are in English.

🔧 Technical Details

The model is based on the English BERT-Small architecture and has 29 million parameters. It is trained using a two-step process with specific loss functions and negative mining strategies. The output dimensions are reduced to 256 using an additional dense layer.

📄 License

No license information is provided in the original document.