Simonc-44/Cygnis-Alpha-2-8B-v0.2

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Mar 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Cygnis-Alpha-2 8B v0.2 by Simonc-44 is an 8 billion parameter Llama 3.1-based language model with merged weights, offering independent execution. It integrates a systematic Chain-of-Thought reasoning process to enhance logical consistency and reduce hallucinations, performing well in both French and English. Optimized for structured role-based interactions via native ChatML support, this model is designed for robust reasoning tasks.

Loading preview...

Cygnis-Alpha-2 8B v0.2 Overview

Cygnis-Alpha-2 8B v0.2, developed by Simonc-44, is a standalone 8 billion parameter model based on the Llama 3.1 architecture. Unlike its predecessor, this version features merged weights, eliminating the need for a separate base model. A core differentiator is its integrated Chain-of-Thought (CoT) reasoning process, designed to improve logical consistency and significantly reduce hallucinations in generated outputs. This makes it particularly effective for tasks requiring structured thought.

Key Capabilities & Features

  • Independent Execution: Full merged weights allow standalone operation.
  • Enhanced Reasoning: Implements CoT processing for logical consistency and reduced hallucinations.
  • Multilingual Performance: Optimized for both French and English language tasks.
  • Native ChatML Support: Designed for structured, role-based conversational interactions.
  • Context Window: Supports an 8192-token context window.

Performance & Quantizations

Benchmarks indicate strong performance compared to similar-sized models, with estimated scores of 45.8 on GSM8K, 61.2 on IFEval, and 52.4 on MMLU. The model is available in various GGUF quantizations, from FP16 to Q2_K, ensuring the reasoning engine's integrity is maintained even at lower bitrates. The Q4_K_M quantization is recommended for most users, offering a balance of quality and efficiency.

Instruction Format

Cygnis-Alpha-2 8B v0.2 uses the ChatML format, including a specific <|im_thought|> token for internal reasoning, which is crucial for leveraging its CoT capabilities.

Finetunes
3 models