Sorawiz/MistralSmall-Creative-24B-Realist

TEXT GENERATIONConcurrency Cost:2Model Size:24BQuant:FP8Ctx Length:32kPublished:Mar 3, 2025Architecture:Transformer0.0K Cold

Sorawiz/MistralSmall-Creative-24B-Realist is a 24 billion parameter merged language model, created by Sorawiz using the TIES merge method with a base of ReadyArt/Forgotten-Safeword-24B-V2.2 and several other models. This highly conversational AI model excels in dialogue generation. It is optimized for producing engaging and interactive text, though it requires careful system prompting to maintain coherence over extended responses.

Loading preview...

Model Overview

Sorawiz/MistralSmall-Creative-24B-Realist is a 24 billion parameter language model developed by Sorawiz. It is a product of a sophisticated merge operation, utilizing the TIES merge method with ReadyArt/Forgotten-Safeword-24B-V2.2 as its foundational base. This model integrates components from multiple other specialized models, including Darkknight535/WinterEngine-24B-Instruct, TroyDoesAI/BlackSheep-24B, and OddTheGreat/Apparatus_24B, among others, to combine their strengths.

Key Capabilities

  • Highly Conversational: Designed to excel in dialogue-based interactions, making it suitable for chat applications and interactive content generation.
  • Merge Architecture: Built using the mergekit tool and the TIES method, indicating a strategic combination of diverse model capabilities.
  • Flexible Chat Templates: Supports both Mistral Instruct and ChatML formats for prompt engineering, offering adaptability for various integration needs.

Considerations for Use

  • Coherence Management: While strong in dialogue, the model may exhibit incoherence in longer responses without a well-designed system prompt. Users should focus on crafting effective system instructions to guide its output.
  • GGUF Availability: Quantized versions (GGUF) are available, thanks to mradermacher, providing options for efficient deployment on various hardware configurations. These include static and Imatrix quants for optimized performance.