ShyliaSafetensors/TritonModelStock-V1-24B

TEXT GENERATIONPricing:Input $0.7 / Cached $0.04 / Output $1.16Concurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 10, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

TritonModelStock-V1-24B is a 24 billion parameter language model developed by ShyliaSafetensors. It was created using the Model Stock merge method with a filter-wise approach, combining five specialized 24B role-play models. This method maintains structural stability while blending the capabilities of its constituent models. The model is based on Magidonia-24B-v4.3-heretic and is designed for role-play scenarios.

Loading preview...

TritonModelStock-V1-24B Overview

TritonModelStock-V1-24B is a 24 billion parameter language model developed by ShyliaSafetensors. This model is notable for its creation via the Model Stock merge method using Mergekit, specifically employing a filter-wise approach. This technique allows for the stable combination of multiple specialized models, blending their unique characteristics while preserving structural integrity.

Key Capabilities & Technical Details

  • Parameter Count: 24 billion parameters.
  • Merge Method: Utilizes the Model Stock method, which is designed to merge models effectively by combining their strengths.
  • Filter-Wise Merging: Enabled to ensure structural stability during the merging process, which is crucial when blending diverse models.
  • Base Model: Built upon Magidonia-24B-v4.3-heretic.
  • Data Type: Operates in bfloat16 precision.
  • Constituent Models: The model is a blend of five distinct 24B role-play models:
    • MS3.2-PaintedFantasy-v2-24B
    • Cydonia-24B-Heretic-v4
    • Dans-PersonalityEngine-V1.3.0-24b
    • Magidonia-24B-v4.3-heretic
    • Eurydice-24b-v3.5

Intended Use Cases

This model is specifically designed for role-play applications, leveraging the combined strengths of its merged components. Its unique merging strategy aims to provide a robust and versatile foundation for generating diverse and engaging role-play scenarios. Users should employ the Mistral V7-Tekken template for optimal performance.

Finetunes
2 models