jina-embedding-v3-m2v-1024

alikia2x/jina-embedding-v3-m2v-1024

This Model2Vec model is a distilled version of the jinaai/jina-embeddings-v3 Sentence Transformer. It uses static embeddings, allowing text embeddings to be computed orders of magnitude faster on both GPU and CPU. It is designed for applications where computational resources are limited or where real-time performance is critical.

Installation

Install model2vec using pip:

pip install model2vec

Usage

Via model2vec

Load this model using the from_pretrained method:

from model2vec import StaticModel

# Load a pretrained Model2Vec model
model = StaticModel.from_pretrained("alikia2x/jina-embedding-v3-m2v-1024")

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

Via sentence-transformers

pip install sentence-transformers

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("alikia2x/jina-embedding-v3-m2v-1024")

# embedding:
# array([[ 1.1825741e-01, -1.2899181e-02, -1.0492010e-01, ...,
#          1.1131058e-03,  8.2779792e-04, -7.6874542e-08]],
#       shape=(1, 1024), dtype=float32)
embeddings = model.encode(["Hello"])

Via ONNX

pip install onnxruntime transformers

You need to download onnx/model.onnx in this repository first.

import onnxruntime
from transformers import AutoTokenizer
import numpy as np

tokenizer_model = "alikia2x/jina-embedding-v3-m2v-1024"
onnx_embedding_path = "path/to/your/model.onnx"

texts = ["Hello"]
tokenizer = AutoTokenizer.from_pretrained(tokenizer_model)
session = onnxruntime.InferenceSession(onnx_embedding_path)

inputs = tokenizer(texts, add_special_tokens=False, return_tensors="np")
input_ids = inputs["input_ids"]
lengths = [len(seq) for seq in input_ids[:-1]]
offsets = [0] + np.cumsum(lengths).tolist()
flattened_input_ids = input_ids.flatten().astype(np.int64)

inputs = {
    "input_ids": flattened_input_ids,
    "offsets": np.array(offsets, dtype=np.int64),
}

outputs = session.run(None, inputs)
embeddings = outputs[0]
embeddings = embeddings.flatten()

# [ 1.1825741e-01 -1.2899181e-02 -1.0492010e-01 ...  1.1131058e-03
#   8.2779792e-04 -7.6874542e-08]
print(embeddings)

Note: A quantized (INT8) version of this model is also available, offering reduced memory usage with minimal performance impact. Simply replace onnx/model.onnx with the onnx/model_INT8.onnx file. Our testing shows less than a 1% drop in the F1 score on a real down-stream task.

How it works

Model2vec creates a small, fast, and powerful model that outperforms other static embedding models by a large margin on all tasks we could find, while being much faster to create than traditional static embedding models such as GloVe. Best of all, you don't need any data to distill a model using Model2Vec.

It works by passing a vocabulary through a sentence transformer model, then reducing the dimensionality of the resulting embeddings using PCA, and finally weighting the embeddings using zipf weighting. During inference, we simply take the mean of all token embeddings occurring in a sentence.

Additional Resources

• All Model2Vec models on the hub
• Model2Vec Repo
• Model2Vec Results
• Model2Vec Tutorials

Library Authors

Model2Vec was developed by the Minish Lab team consisting of Stephan Tulkens and Thomas van Dongen.

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},
}