Liontix/ruby-9b

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 13, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Liontix/ruby-9b is an experimental 9 billion parameter model developed by Liontix, fine-tuned from armand0e/Qwen3.5-9B-Opus-Agent. This model was trained using Supervised Fine-Tuning (SFT) on Gemini 3 Flash Preview responses, specifically masking the Chain of Thought (CoT) during training. It is designed for experimental use in exploring SFT techniques with masked CoT, rather than production applications.

Loading preview...

Overview

Liontix/ruby-9b is an experimental 9 billion parameter language model developed by Liontix, fine-tuned from armand0e/Qwen3.5-9B-Opus-Agent. This model explores a specific Supervised Fine-Tuning (SFT) methodology where the Chain of Thought (CoT) part of Gemini 3 Flash Preview responses was masked during training.

Key Characteristics

  • Experimental Nature: Explicitly stated as not meant for production or serious use, serving as a testbed for SFT techniques.
  • Training Methodology: Utilizes SFT on Gemini 3 Flash Preview responses with a unique approach of masking the CoT during the training process.
  • Future Development: The developer plans to release more models using a fine-tuned base model with raw/synthetic CoT, trained on a larger dataset without masking CoT.
  • Tooling: Leverages unsloth for open-source tooling in its development.

Intended Use

This model is primarily for:

  • Research and Experimentation: Ideal for developers and researchers interested in exploring advanced SFT techniques, particularly those involving Chain of Thought masking.
  • Understanding SFT Impact: Provides a basis for understanding how masking CoT during SFT affects model behavior and performance.

Note: This model is explicitly marked as experimental and not suitable for production environments.

Finetunes
2 models