cstr/jina-reranker-v2-base-multilingual-GGUF

jina-reranker-v2-base-multilingual GGUF

GGUF format of jinaai/jina-reranker-v2-base-multilingual for use with CrispEmbed.

Jina Reranker v2 Base Multilingual. Cross-encoder reranker for 100+ languages. Post-LN, NomicBERT-like layout (mixer.Wqkv + GELU FFN). Use with crispembed_rerank().

Files

File	Quantization	Size
jina-reranker-v2-base-multilingual.gguf	F32	1068 MB
jina-reranker-v2-base-multilingual-q8_0.gguf	Q8_0	282 MB
jina-reranker-v2-base-multilingual-q4_k.gguf	Q4_K	242 MB

Parity vs HuggingFace reference

Cosine similarity vs the upstream sentence-transformers reference on a fixed test set (text):

Quant	Text
q8_0	0.9997
q4_k	0.9981

Quick Start

# Download
huggingface-cli download cstr/jina-reranker-v2-base-multilingual-GGUF jina-reranker-v2-base-multilingual.gguf --local-dir .

# Run with CrispEmbed
./crispembed -m jina-reranker-v2-base-multilingual.gguf "Hello world"

# Or with auto-download
./crispembed -m jina-reranker-v2-base-multilingual "Hello world"

Model Details

Property	Value
Architecture	Jina v2 (XLM-R variant)
Parameters	278M
Embedding Dimension	768
Layers	12
Pooling	CLS
Tokenizer	SentencePiece
Base Model	jinaai/jina-reranker-v2-base-multilingual

Verification

Verified bit-identical to HuggingFace sentence-transformers (cosine similarity >= 0.999 on test texts).

Usage with CrispEmbed

CrispEmbed is a lightweight C/C++ text embedding inference engine using ggml. No Python runtime, no ONNX. Supports BERT, XLM-R, Qwen3, and Gemma3 architectures.

# Build CrispEmbed
git clone https://github.com/CrispStrobe/CrispEmbed
cd CrispEmbed
cmake -S . -B build && cmake --build build -j

# Encode
./build/crispembed -m jina-reranker-v2-base-multilingual.gguf "query text"

# Server mode
./build/crispembed-server -m jina-reranker-v2-base-multilingual.gguf --port 8080
curl -X POST http://localhost:8080/v1/embeddings \
    -d '{"input": ["Hello world"], "model": "jina-reranker-v2-base-multilingual"}'

Credits

• Original model: jinaai/jina-reranker-v2-base-multilingual
• Inference engine: CrispEmbed (ggml-based)
• Conversion: convert-bert-embed-to-gguf.py