redis-vl-dotnet

ONNX Reranker

RedisVL.Rerankers.Onnx provides OnnxTextReranker, an ITextReranker implementation that scores query-plus-document pairs locally with ONNX Runtime.

Package contents

OnnxTextReranker for local cross-encoder reranking
OnnxRerankerOptions for model asset and runtime configuration
OnnxRuntimeSessionOptions for ONNX Runtime session tuning
OnnxRerankerPackage as the package marker type

Local model assets

This package does not bundle model files. Applications must provide:

a local model.onnx file
a local Hugging Face tokenizer.json file

The reference workflow assumes a BERT-style cross-encoder tokenizer shape:

[CLS] query [SEP] document [SEP]
input_ids
attention_mask
token_type_ids when the model expects them

Request options

OnnxRerankerOptions currently supports:

ModelPath for the local ONNX model file
TokenizerPath for the local tokenizer definition
MaxSequenceLength with a default of 512
ScoreThreshold when the application wants to drop low-scoring matches after inference
SessionOptions for local ONNX Runtime tuning such as graph optimization, execution mode, and thread counts

ScoreThreshold is optional because score calibration can vary across exported cross-encoder models.

Example workflow

/examples/OnnxRerankerExample shows the expected integration pattern:

load an initial candidate set
print the original candidate order
rerank those candidates with OnnxTextReranker
print the local ONNX-adjusted order

Run it from the repository root after setting local asset paths:

export ONNX_RERANKER_MODEL_PATH=/path/to/model.onnx
export ONNX_RERANKER_TOKENIZER_PATH=/path/to/tokenizer.json
dotnet run --project examples/OnnxRerankerExample/OnnxRerankerExample.csproj

Validation references

tests/RedisVL.Tests/Rerankers/OnnxTextRerankerTests.cs covers validation, ordering, TopN, and threshold behavior
tests/RedisVL.Tests/Rerankers/OnnxTextRerankerSmokeTests.cs is the local-asset smoke test gate