Naphula/SpaceBound-24B-v1
Naphula/SpaceBound-24B-v1 is an experimental 24 billion parameter model based on the MistralForCausalLM architecture, created by Naphula as a hybrid fork expansion of the WeirdCompound/BereavedCompound series. This model is a complex, multi-stage merge of numerous specialized models, including those focused on roleplay, prompt adherence, and various narrative styles, with a context length of 32768 tokens. It is primarily intended for archival and testing purposes to evaluate the performance of its intricate merging pipeline. Its unique composition aims to explore the effects of combining diverse model characteristics.
Loading preview...
SpaceBound-24B-v1 Overview
Naphula/SpaceBound-24B-v1 is an experimental 24 billion parameter language model built upon the MistralForCausalLM architecture. It represents a complex, multi-stage merge (slerp, nuslerp, flux) of over 20 distinct base models, including TheDrummer/Cydonia-24B-v4.3 and anthracite-core/Mistral-Small-3.2-24B-Instruct-2506-Text-Only, among others. This intricate merging process, which took over 12 hours, aims to combine a wide array of specialized capabilities.
Key Characteristics
- Hybrid Merge Architecture: Utilizes a unique
model_stock→slerp→nuslerp→fluxpipeline, integrating diverse models. - Broad Specialization: Incorporates models designed for specific traits such as roleplay (
aixonlab/Eurydice-24b-v3.5,ReadyArt/MS3.2-The-Omega-Directive-24B-Unslop-v2.1), prompt adherence (PocketDoc/Dans-PersonalityEngine-V1.3.0-24b), randomness (TheDrummer/Rivermind-24B-v1), and various narrative styles (e.g., 'animu', 'adventure', 'dragons'). - Experimental Nature: Explicitly stated as an archival and testing model, acknowledging potential performance degradation compared to its base models.
Intended Use
This model is primarily for archival and testing purposes to evaluate the outcomes of its complex, multi-stage merging methodology. Developers interested in exploring the effects of combining numerous specialized models through advanced merging techniques may find it useful for research and experimentation.