🪲 red-beetle-small-v1 Model Card

[!TIP] Beetles are some of the most diverse and interesting creatures on Earth. They are found in every environment, from the deepest oceans to the highest mountains. They are also known for their ability to adapt to a wide range of habitats and lifestyles. They are small, fast and powerful!

The beetle series of models are made as good starting points for Static Embedding training (via TokenLearn or Fine-tuning), as well as decent Static Embedding models. Each beetle model is made to be an improvement over the original M2V_base_output model in some way, and that's the threshold we set for each model (except the brown beetle series, which is the original model).

This model has been distilled from mixedbread-ai/mxbai-embed-2d-large-v1, with PCA at 384 dimensions and (Zipf)^3 re-weighting. It outperforms the original M2V_base_output model in all tasks.

Version Information

• red-beetle-base-v0: The original model, without using PCA or Zipf. The lack of PCA and Zipf also makes this a decent model for further training.
• red-beetle-base-v1: The original model, with PCA at 1024 dimensions and (Zipf)^3 re-weighting.
• red-beetle-small-v1: A smaller version of the original model, with PCA at 384 dimensions and (Zipf)^3 re-weighting.
• red-beetle-base-v1.1: The original model, with PCA at 1024 dimensions, Zipf and SIF re-weighting, learnt from a subset of the FineWeb-Edu sample-10BT dataset.
• red-beetle-small-v1.1: A smaller version of the original model, with PCA at 384 dimensions, Zipf and SIF re-weighting, learnt from a subset of the FineWeb-Edu sample-10BT dataset.

Installation

Install model2vec using pip:

pip install model2vec

Usage

Load this model using the from_pretrained method:

from model2vec import StaticModel

# Load a pretrained Model2Vec model
model = StaticModel.from_pretrained("bhavnicksm/red-beetle-small-v1")

# Compute text embeddings
embeddings = model.encode(["Example sentence"])

Read more about the Model2Vec library here.

Comparison with other models

Coming soon...

Acknowledgements

This model is made using the Model2Vec library. Credit goes to the Minish Lab team for developing this library.

Citation

Please cite the Model2Vec repository if you use this model in your work.

@software{minishlab2024model2vec,
  authors = {Stephan Tulkens, Thomas van Dongen},
  title = {Model2Vec: Turn any Sentence Transformer into a Small Fast Model},
  year = {2024},
  url = {https://github.com/MinishLab/model2vec},
}