OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Bge M3 Onnx

Bge M3 Onnx is an open-source repository demonstrating conversion of the BGE-M3 multilingual embedding model and tokenizer to ONNX, with sample inference implementations in C#, Java and Python. It generates dense, sparse and ColBERT embeddings locally, and the repository describes offline operation and optional CUDA acceleration.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Bge M3 Onnx — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when an application needs locally generated BGE-M3 embeddings in ONNX format, particularly when it needs more than dense vectors or must operate offline.

Who it is for

  • Developers adding multilingual text embeddings to applications written in C#, Java or Python.
  • Teams building local or offline document-embedding and retrieval pipelines.

Documented uses

  • Generate dense vectors for semantic search indexing.
  • Produce sparse lexical weights and ColBERT vectors for hybrid retrieval workflows.
  • Convert BGE-M3 into ONNX model files for local inference.
  • Compare Python-generated reference embeddings with .NET and Java outputs using the repository's optional test scripts.
Before you choose

Requirements & limits

The public GitHub repository specifies an Apache-2.0 license. The README directs users to obtain ONNX model and tokenizer files from repository releases or generate them with its notebook.

  • The repository supplies samples for Python, C# and Java; select tooling and runtime setup for the language you use.
  • Model and tokenizer files are a separate setup step: obtain the release archive or create them with the conversion notebook.
  • The output includes dense, sparse and ColBERT embeddings, rather than only a single vector representation.
  • CUDA acceleration has distinct setup requirements: Python uses an ONNX Runtime CUDA package, while C# and Java require CUDA and cuDNN installed separately.
  • Optional cross-language comparison scripts require Python, .NET, Java and Maven.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Generate embeddings with a repository sample

Prepare the ONNX model and tokenizer, then use a language sample to generate embeddings from text.

Before you start

  • A clone of the repository
  • The ONNX model and tokenizer, obtained from the repository release archive or generated with the conversion notebook
  • Python, .NET or Java tooling appropriate to the sample; cross-language comparison scripts additionally require Maven

Steps

  1. Clone the repository.

  2. Download onnx.zip from the repository releases, or run bge-m3-to-onnx.ipynb to create the model and tokenizer files in /onnx.

  3. Use a sample in /samples/python, /samples/dotnet or /samples/java with the model and tokenizer paths.

  4. Call the sample's embedding method on text to obtain dense, sparse and ColBERT outputs.

Expected outcome

The sample produces all three BGE-M3 embedding types for supplied text.

Guide limitations

  • The repository provides sample implementations rather than a hosted embedding service.
  • CUDA acceleration requires additional runtime and CUDA/cuDNN setup; requirements differ between Python and C#/Java.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
Bge M3 OnnxGenerate dense vectors for semantic search indexing.Linux, macOS, Self-hosted, WindowsThe public GitHub repository specifies an Apache-2.0 license. The README directs users to obtain ONNX model and tokenizer files from repository releases or generate them with its notebook.
Pulse Of MotionEstimate a video's physical frame rate from its visual dynamics.LinuxThe repository is public. Its README describes a released checkpoint that downloads from Hugging Face on first inference, with a manual download option. The strongest internal model is not fully released; the README directs interested users to contact the project about access. No license information is provided in the repository evidence.
Neural Image AssessmentTrain the model using AVA images and rating annotations.Cloud, Linux, macOS, Windows
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
yuniko-software/bge-m3-onnx · Official homepage title · retained in its source language
How we review sourcesSuggest a correction