README.md

metadata

library_name: sentence-transformers
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - sentence-similarity
  - feature-extraction
  - dense
  - arabic

NeoAraBERT-STS

Sentence-transformers model for Arabic semantic textual similarity.

Usage

pip install -U sentence-transformers torch

import torch
from sentence_transformers import SentenceTransformer

model_name = "U4RASD/NeoAraBERT-STS"

finetuned_model = SentenceTransformer(
    model_name,
    model_kwargs={"trust_remote_code": True, "torch_dtype": torch.float32},
    tokenizer_kwargs={"trust_remote_code": True},
    config_kwargs={"trust_remote_code": True},
)
finetuned_model.max_seq_length = 512

sentences = [
    "التقارير بدأت تصل في وقت متأخر من هذا العام ويتم مراجعتها",
    "يتم مراجعة التقارير في أواخر هذا العام.",
    "لم يكن هناك تقارير هذا العام على الإطلاق.",
]

embeddings = finetuned_model.encode(sentences)
similarities = finetuned_model.similarity(embeddings, embeddings)

print(embeddings.shape)
print(similarities)

Model Type

• Model type: Sentence Transformer
• Task: Sentence similarity / semantic textual similarity
• Language: Arabic
• Embedding size: 768
• Max sequence length: 512
• Similarity function: Cosine similarity