NIFE GTE-ModernBERT-Base

This is a NIFE model finetuned from Alibaba-NLP/gte-modernbert-base. It is fully aligned with its base model Alibaba-NLP/gte-modernbert-base, and can be used to perform Inference Free querying using an index made by this model. It can also be used in standalone mode.

Model Details

Model Description

Model Type: Static Model
Base model: Alibaba-NLP/gte-modernbert-base
Output Dimensionality: 768 dimensions

Full Model Architecture

SentenceTransformer(
  (0): TrainableStaticEmbedding(
    (embedding): EmbeddingBag(100003, 768, mode='mean')
  )
  (1): Normalize()
)

Usage

Direct Usage (NIFE)

First install the NIFE library

pip install -U pynife

Then you can run the model as follows:

from nife import load_as_router

model = load_as_router("stephantulkens/NIFE-gte-modernbert-base")

query = "What is the capital of France?"
query_embeddings = model.encode_query(query)

# Five locales near France
index_doc = model.encode_document(["Paris is the largest city in France", "Lyon is pretty big", "Antwerp is really great, and in Belgium", "Berlin is pretty gloomy in winter", "France is a country in Europe"])

similarity = model.similarity(query_vec, index_doc)
print(similarity)
# It correctly retrieved the document containing the statement about paris.
# tensor([[0.7065, 0.5012, 0.3596, 0.2765, 0.6648]])

Training Details

Citation

BibTeX

@software{Tulkens2025pyNIFE,
  author       = {St\'{e}phan Tulkens},
  title        = {pyNIFE: nearly inference free embeddings in python},
  year         = {2025},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.17512919},
  url          = {https://github.com/stephantulkens/pynife},
  license      = {MIT},
}

stephantulkens/NIFE-gte-modernbert-base