Roxas13/e5-small-mlx

AEGIS multilingual-e5-small (FP16, MLX)

FP16, MLX-format build of intfloat/multilingual-e5-small, used on-device by the AEGIS app for RAG memory retrieval (384-dim sentence embeddings). This is a format conversion of the original (PyTorch → MLX, FP16) — no fine-tuning and no architecture change.

License & attribution

This repository redistributes a derivative of intfloat/multilingual-e5-small, which is released under the MIT License. The original copyright and MIT license terms are retained and passed along to all recipients.

• Source model: https://huggingface.co/intfloat/multilingual-e5-small
• License: MIT

Modifications relative to the base model: weights converted to MLX format at FP16 precision. Output is 384-dim with mean pooling + L2 normalization. No weights were fine-tuned.

Usage notes

E5 requires input prefixes for good retrieval accuracy: query: for search queries and passage: for stored documents.