Snowflake Arctic Embed M v2.0. Multilingual retrieval embeddings, 768-dimensional, 8192-token context. Queries take the prefix "query: "; documents take no prefix.

Files

File	Quantization	Size
arctic-embed-m-v2-q4_k.gguf	Q4_K	261 MB
arctic-embed-m-v2-q8_0.gguf	Q8_0	315 MB
arctic-embed-m-v2.gguf	F32	1170 MB

Quick Start

bash

# Download
huggingface-cli download cstr/arctic-embed-m-v2-GGUF arctic-embed-m-v2-q4_k.gguf --local-dir .

# Run with CrispEmbed
./crispembed -m arctic-embed-m-v2-q4_k.gguf "Hello world"

# Or with auto-download
./crispembed -m arctic-embed-m-v2 "Hello world"

Model Details

Property	Value
Architecture	GTE v1.5 (RoPE, GeGLU, post-LN)
Parameters	305M
Embedding Dimension	768
Layers	12
Pooling	CLS
Tokenizer	SentencePiece
Base Model	Snowflake/snowflake-arctic-embed-m-v2.0

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.

bash

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

# Encode
./build/crispembed -m arctic-embed-m-v2-q4_k.gguf "query text"

# Server mode
./build/crispembed-server -m arctic-embed-m-v2-q4_k.gguf --port 8080
curl -X POST http://localhost:8080/v1/embeddings \
    -d '{"input": ["Hello world"], "model": "arctic-embed-m-v2"}'

Credits

Original model: Snowflake/snowflake-arctic-embed-m-v2.0
Inference engine: CrispEmbed (ggml-based)
Conversion: convert-bert-embed-to-gguf.py