voice-e5-small-int8

On-device meaning search for the «Слышно» voice recorder app (smart notes: «Ask the archive», «Similar recordings»). Vectors are computed on the phone; nothing is sent anywhere.

model.int8.onnx - intfloat/multilingual-e5-small, ONNX dynamic int8 quantization by Xenova, unchanged.
vocab.bin - the model's SentencePiece Unigram vocabulary (from tokenizer.json) repacked into a compact binary: "SPU1", int32 count, then per piece float32 score, uint8 byte length, UTF-8 bytes.

Usage: prefix query: / passage: , mean pooling over tokens, L2 normalisation.

License: MIT (same as the base model). © Microsoft Corporation / intfloat.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for krut42/voice-e5-small-int8

Base model

intfloat/multilingual-e5-small

Quantized

(294)

this model