Cygnis-Alpha-2 8B v0.3: Sovereign Reasoning Engine

Cygnis-Alpha-2 8B v0.3, developed by Simonc-44, is a fine-tuned version of the Llama 3.1 8B model, engineered for stability and transparent reasoning. It integrates a unique native Chain-of-Thought (CoT) mechanism, utilizing a dedicated <|im_thought|> token to structure its internal thought processes, enhancing logical consistency.

Key Capabilities

• Native Chain-of-Thought (CoT): Employs a specific token for internal reasoning, promoting structured and transparent outputs.
• Bilingual Excellence: Fully fine-tuned for high performance in both French and English.
• Production-Ready Stability: Addresses critical architecture mismatches, ensuring reliable deployment with tools like Ollama and Llama.cpp.

Architecture & Training

This model was trained using Unsloth on two specialized datasets:

• Cygnis-Alpha2-Instruct-Mix for multi-step reasoning tasks.
• Cygnis-Identity-SFT for establishing a sovereign persona and ethical alignment.

Good for

• Applications requiring robust, logically structured responses.
• Bilingual (French/English) conversational AI and content generation.
• Deployments needing stable and compatible models with Ollama and Llama.cpp.
• Use cases where transparent reasoning and ethical alignment are paramount.