ONNX Model Files

The model files are not stored in git (too large). They are downloaded automatically by maven-download-plugin during the MCP module build, or manually:

curl -L -o model_quantized.onnx "https://huggingface.co/onnx-community/granite-embedding-small-english-r2-ONNX/resolve/main/onnx/model_quantized.onnx"
curl -L -o model_quantized.onnx_data "https://huggingface.co/onnx-community/granite-embedding-small-english-r2-ONNX/resolve/main/onnx/model_quantized.onnx_data"

• Model: ibm-granite/granite-embedding-small-english-r2 (ONNX Q8 quantized)
• Dimensions: 384
• Max sequence length: 8,192 tokens (capped at 2,048 in OnnxEmbeddingProvider; override with -Dembedding.maxLength)
• Size: ~52 MB (graph: 0.6 MB + weights: 49 MB)
• Architecture: ModernBERT (12 layers, 12 heads, 47M params)

Reranker (cross-encoder)

Downloaded by the MCP module build as reranker_quantized.onnx + reranker-tokenizer.json:

curl -L -o reranker_quantized.onnx "https://huggingface.co/Xenova/ms-marco-MiniLM-L-6-v2/resolve/main/onnx/model_quantized.onnx"
curl -L -o reranker-tokenizer.json "https://huggingface.co/Xenova/ms-marco-MiniLM-L-6-v2/resolve/main/tokenizer.json"

• Model: cross-encoder/ms-marco-MiniLM-L-6-v2 (ONNX Q8 quantized)
• Purpose: reranks top hybrid-search candidates in LuceneSearchService.search() (see OnnxReranker)
• Max sequence length: 512 tokens (query + passage pair, longest-first truncation)
• Size: ~23 MB