Blablablab / multilingual-style-representation-Llama-3.2

Multilingual Style Representation based on meta-llama/Llama-3.2-1B

This is the Style Representation model, presented in Leveraging Multilingual Training for Authorship Representation: Enhancing Generalization across Languages and Domains .

The Style Representation model encodes documents written by the same author as nearby vectors in the embedding space. The model can be used for authorship attribution, style similarity, machine-generated text detection, and more.

For training and evaluation code, refer to our repository here .

For the Style Representation model based on XLM-RoBERTa, refer to Blablablab/multilingual-style-representation .

Model Details

• Model Type: Sentence Transformer
• Base model: meta-llama/Llama-3.2-1B
• Maximum Sequence Length: 131072 tokens
• Output Dimensionality: 2048 dimensions
• Similarity Function: Cosine Similarity

Usage

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("Blablablab/multilingual-style-representation-Llama-3.2")
# Run inference
sentences = [
    'The weather is lovely today.',
    "It's so sunny outside!",
    'He drove to the stadium.',
]
embeddings = model.encode(sentences)
print(embeddings.shape)
# [3, 2048]

# Get the similarity scores for the embeddings
similarities = model.similarity(embeddings, embeddings)
print(similarities.shape)
# [3, 3]