Instructions to use JasonYANG170/multilingual-e5-small-ONNX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers.js
How to use JasonYANG170/multilingual-e5-small-ONNX with Transformers.js:
// npm i @huggingface/transformers import { pipeline } from '@huggingface/transformers'; // Allocate pipeline const pipe = await pipeline('feature-extraction', 'JasonYANG170/multilingual-e5-small-ONNX');
multilingual-e5-small-ONNX
Independently converted from original weights at intfloat/multilingual-e5-small@614241f622f53c4eeff9890bdc4f31cfecc418b3.
No third-party ONNX weights were used. Original authorship and license are retained;
see UPSTREAM_MODEL_CARD.md and included license files where supplied by upstream.
Runtime
Transformers.js 4.0.1, dtype: 'q8'. Browser WASM validation passed on the conversion
workstation; this does not certify performance on a 16 GB integrated-GPU device.
Use session_options: {graphOptimizationLevel: 'basic'} for Marian translation.
E5 requires query: / passage: prefixes, mean pooling and normalization (384 dimensions).
Marian tokenizer uses the original source-language SentencePiece scores.
Reproducibility
See conversion.json for source revision, checksums and validation results. The conversion/ directory contains the conversion scripts, source lock and dependency lock. FP32 reference graphs are retained locally; this repository ships the validated q8 graphs.
- Downloads last month
- 152
Model tree for JasonYANG170/multilingual-e5-small-ONNX
Base model
intfloat/multilingual-e5-small