bge-m3_dynamic_int8_onnx

---

language:

• en
• zh license: mit tags:
• sentence-transformers
• feature-extraction
• sentence-similarity
• transformers
• onnx
• quantized
• reranker
• bge library_name: onnxruntime pipeline_tag: feature-extraction base_model: BAAI/bge-reranker-v2-m3 model-index:
• name: bge-m3-onnx-int8 results:
  • task: type: information-retrieval name: Information Retrieval metrics:
    • type: performance_retention value: 98 name: Performance Retention (%)
    • type: model_size_reduction value: 75 name: Model Size Reduction (%)

---

BGE-M3 ONNX INT8 Quantized Model

This is an ONNX version of the BAAI/bge-reranker-v2-m3 model, optimized with dynamic INT8 quantization for efficient inference.

Model Description

• Base Model: BAAI/bge-reranker-v2-m3
• Model Type: Sentence Transformer / Reranker
• Quantization: Dynamic INT8
• Framework: ONNX Runtime
• Model Size: ~560MB (75% reduction from original ~2.2GB)
• Performance: Maintains 98%+ accuracy compared to the original model

Key Features

• ✅ Efficient: 75% model size reduction with minimal accuracy loss
• ✅ Fast Inference: Optimized for CPU and GPU acceleration
• ✅ Cross-Platform: Compatible with ONNX Runtime on multiple platforms
• ✅ Production Ready: Suitable for deployment in resource-constrained environments

Usage

Prerequisites

pip install onnxruntime transformers numpy