guymorganb/multilingual-e5-base-int8

multilingual-e5-base — 8-bit ONNX mirror for Text Embeddings Inference

A mirror of intfloat/multilingual-e5-base at revision d128750597153bb5987e10b1c3493a34e5a4502a whose onnx/model.onnx is the upstream repository's own onnx/model_qint8_avx512_vnni.onnx (279 MB, 8-bit), renamed so that Text Embeddings Inference's ONNX backend loads it by default. Every other file is the upstream file, unchanged. Made 2026-10-29 so the model fits a 1 GB instance (measured: 684 MiB idle, 736 MiB under load) where the 32-bit file (1.1 GB) does not. Same 768 dimensions, same query: / passage: prefixes, same 512-token input. License: MIT, as upstream.