cstr/pixie-rune-v1-GGUF

pixie-rune-v1 GGUF

GGUF format of telepix/PIXIE-Rune-v1.0 for use with CrispEmbed and Ollama.

Files

File	Quantization	Size
pixie-rune-v1-q4_k.gguf	Q4_K	0 MB
pixie-rune-v1-q8_0.gguf	Q8_0	0 MB
pixie-rune-v1.gguf	F32	0 MB

Recommended: Q8_0 for quality (cos vs HF: cross-lingual OK), Q4_K for size (cross-lingual OK).

Quick Start

CrispEmbed

./crispembed -m pixie-rune-v1 "Hello world"
./crispembed-server -m pixie-rune-v1 --port 8080

Ollama (with CrispStrobe fork)

echo "FROM pixie-rune-v1-q8_0.gguf" > Modelfile
ollama create pixie-rune-v1 -f Modelfile
curl http://localhost:11434/api/embed -d '{"model":"pixie-rune-v1","input":["Hello world"]}'

Python (CrispEmbed)

from crispembed import CrispEmbed
model = CrispEmbed("pixie-rune-v1-q8_0.gguf")
vectors = model.encode(["Hello world", "Goodbye world"])

Model Details

Property	Value
Architecture	XLM-R
Parameters	560M
Embedding Dimension	1024
Layers	24
Pooling	CLS
Tokenizer	SentencePiece
Language	multilingual
Q8_0 vs HuggingFace	cross-lingual OK
Q4_K vs HuggingFace	cross-lingual OK

Server API

CrispEmbed server supports four API dialects:

• POST /embed -- native
• POST /v1/embeddings -- OpenAI-compatible
• POST /api/embed -- Ollama-compatible
• POST /api/embeddings -- Ollama legacy

Credits

• Original model: telepix/PIXIE-Rune-v1.0
• Inference: CrispEmbed (MIT, ggml-based)