GATE-AraBert-V1

This is GATE | General Arabic Text Embedding trained using SentenceTransformers in a multi-task setup. The system trains on the AllNLI and on the STS dataset.

Downloads last month: 15

Safetensors

Model size

0.1B params

Tensor type

F32

Collection including LLMXperts/GATE-AraBert-v1

GATE Repo

Collection

Models and Dataset for Gate Paper Sumbission • 6 items • Updated Apr 17, 2025

Evaluation results

cosine_pearson on MTEB STS17 (ar-ar)
test set self-reported

82.066
cosine_spearman on MTEB STS17 (ar-ar)
test set self-reported

82.781
euclidean_pearson on MTEB STS17 (ar-ar)
test set self-reported

79.240
euclidean_spearman on MTEB STS17 (ar-ar)
test set self-reported

81.529
main_score on MTEB STS17 (ar-ar)
test set self-reported

82.781
manhattan_pearson on MTEB STS17 (ar-ar)
test set self-reported

78.954
manhattan_spearman on MTEB STS17 (ar-ar)
test set self-reported

81.151
cosine_pearson on MTEB STS22.v2 (ar)
test set self-reported

54.913