cstr/granite-embedding-97m-multilingual-r2-GGUF

granite-embedding-97m-multilingual-r2 GGUF

GGUF format of ibm-granite/granite-embedding-97m-multilingual-r2 for use with CrispEmbed.

IBM Granite Embedding 97M R2. Multilingual (50+ languages), 384-dimensional CLS-pooled, 8192-token context. No query or document prefix.

Files

File	Quantization	Size
granite-embedding-97m-multilingual-r2-f16.gguf	F16	362 MB
granite-embedding-97m-multilingual-r2-q8_0.gguf	Q8_0	106 MB

Parity vs HuggingFace reference

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

Quant	Text
f16	1.0000
q8_0	0.9996

Quick Start

# Download
huggingface-cli download cstr/granite-embedding-97m-multilingual-r2-GGUF granite-embedding-97m-multilingual-r2-f16.gguf --local-dir .

# Run with CrispEmbed
./crispembed -m granite-embedding-97m-multilingual-r2-f16.gguf "Hello world"

# Or with auto-download
./crispembed -m granite-embedding-97m-multilingual-r2 "Hello world"

Model Details

Property	Value
Architecture	ModernBERT (RoPE, local/global attention, SwiGLU, pre-LN)
Parameters	97M
Embedding Dimension	384
Layers	12
Pooling	CLS
Tokenizer	o200k BPE (180k)
Base Model	ibm-granite/granite-embedding-97m-multilingual-r2

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 granite-embedding-97m-multilingual-r2-f16.gguf "query text"

# Server mode
./build/crispembed-server -m granite-embedding-97m-multilingual-r2-f16.gguf --port 8080
curl -X POST http://localhost:8080/v1/embeddings \
    -d '{"input": ["Hello world"], "model": "granite-embedding-97m-multilingual-r2"}'

Credits

• Original model: ibm-granite/granite-embedding-97m-multilingual-r2
• Inference engine: CrispEmbed (ggml-based)
• Conversion: convert-bert-embed-to-gguf.py