cstr/octen-8b-GGUF

octen-8b GGUF

GGUF format of Octen/Octen-Embedding-8B for use with CrispEmbed.

Octen Embedding 8B. Qwen3-based decoder, 4096-dimensional, last-token pooling.

Files

File	Quantization	Size
octen-8b-q4_k.gguf	Q4_K	4362 MB
octen-8b-q5_k.gguf	Q5_K	5190 MB
octen-8b-q8_0.gguf	Q8_0	7674 MB
octen-8b.gguf	F32	28873 MB

Parity vs HuggingFace reference

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

Quant	Text
q4_k	0.9654

Note: below the 0.99 retrieval-quality bar — text: q4_k (0.965). Embeddings are still functionally usable (>0.9 = directionally correct for similarity ranking) but expect small differences in nearest-neighbor results vs the upstream f32 reference.

Quick Start

# Download
huggingface-cli download cstr/octen-8b-GGUF octen-8b-q4_k.gguf --local-dir .

# Run with CrispEmbed
./crispembed -m octen-8b-q4_k.gguf "Hello world"

# Or with auto-download
./crispembed -m octen-8b "Hello world"

Model Details

Property	Value
Architecture	Qwen3
Parameters	8B
Embedding Dimension	4096
Layers	36
Pooling	last-token
Tokenizer	GPT-2 BPE
Base Model	Octen/Octen-Embedding-8B

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 octen-8b-q4_k.gguf "query text"

# Server mode
./build/crispembed-server -m octen-8b-q4_k.gguf --port 8080
curl -X POST http://localhost:8080/v1/embeddings \
    -d '{"input": ["Hello world"], "model": "octen-8b"}'

Credits

• Original model: Octen/Octen-Embedding-8B
• Inference engine: CrispEmbed (ggml-based)
• Conversion: convert-decoder-embed-to-gguf.py