TheDrummer/Cydonia-24B-v3.1

Hugging Face
TEXT GENERATIONConcurrency Cost:2Model Size:24BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 12, 2025Architecture:Transformer0.0K Warm

TheDrummer/Cydonia-24B-v3.1 is a 24 billion parameter language model based on the Mistral v7 Tekken architecture, developed by TheDrummer. Optimized for character-driven roleplay and story generation, it excels at maintaining character consistency and narrative flow. This model supports an optional thinking mode for enhanced reasoning and is designed for users seeking engaging and responsive interactive fiction experiences.

Loading preview...

Cydonia 24B v3.1 Overview

Cydonia 24B v3.1, developed by TheDrummer, is a 24 billion parameter language model built on the Mistral v7 Tekken architecture. It is specifically designed for character-focused roleplay and interactive storytelling, demonstrating strong capabilities in driving narratives and maintaining character consistency.

Key Capabilities

  • Character-Driven Roleplay: Excels at playing characters, driving story progression, and adapting to scenario changes, including "spicy" content when appropriate.
  • Prompt Following: Reliably follows user prompts and formats messages as expected.
  • Optional Reasoning (<think>): Features an optional thinking mode, which, while not automatic, can enhance reasoning capabilities with proper prompt prefill (e.g., <think>). Users note its strong performance even without explicit reasoning prompts.
  • Hardware Friendly: Described as "comfy to use" for users with 24GB of VRAM, offering quick generation and enjoyable output across various hardware configurations.

Good For

  • Interactive Fiction & Roleplay: Ideal for scenarios requiring dynamic character interaction and story generation.
  • Users with 24GB VRAM: Optimized for efficient use on systems with 24GB of GPU memory, providing a balance of performance and resource consumption.
  • Exploratory Prompting: Users can experiment with the <think> prefill for enhanced reasoning, though it requires some prompt wrangling as it's not a special token.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p