Cygnis Alpha 2 8B

v0.3 Stable GGUF Edition

High-precision quantizations of the Sovereign Reasoning Engine.
Optimized for local production and private, high-speed inference.

Quantization Distribution

Quantization	Size	Technical Tag	Availability
f16	16.10 GB	UNCOMPRESSED	Link
Q8_0	8.54 GB	PRO REFERENCE	Link
Q6_K	6.60 GB	HIGH QUALITY	Link
Q4_K_M	4.92 GB	GOLDEN MEAN	Link
Q4_K_S	4.69 GB	BALANCED FAST	Link
IQ4_XS	4.48 GB	IMATRIX SMALL	Link
Q3_K_L / M / S	~4.00 GB	LOW RAM TRIO	View 3 files
Q2_K	3.18 GB	ULTRA LIGHT	Link

Perplexity vs Bitrate

Reference benchmark for llama-3.1-8b quantization curves.

Ollama Integration

FROM ./Cygnis-Alpha-2-8B-v0.3.Q4_K_M.gguf

TEMPLATE """<|im_start|>system
{{ .System }}<|im_end|>
<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
<|im_thought|>
"""
PARAMETER stop "<|im_end|>"

Developed by Simonc-44