FlareRebellion/BereavedCompound-v1.0-24b

Warm
Public
24B
FP8
32768
Nov 19, 2025
Hugging Face
Overview

BereavedCompound-v1.0-24b: A Multi-Stage Merged Model

BereavedCompound-v1.0-24b is a 24 billion parameter language model developed by FlareRebellion, resulting from a complex, multi-stage merging process using mergekit. This model leverages several advanced merge methods, including Model Stock, SLERP, and NuSLERP, to combine the capabilities of multiple specialized base models.

Key Capabilities & Merged Strengths

The model's architecture is built upon ReadyArt/MS3.2-The-Omega-Directive-24B-Unslop-v2.0 as a primary base, integrating diverse functionalities from other models. Notable contributions from its constituent models include:

  • Storytelling and Roleplay: Incorporates aixonlab/Eurydice-24b-v3.5 to enhance narrative generation and character interaction.
  • Prompt Adherence: Benefits from PocketDoc/Dans-PersonalityEngine-V1.3.0-24b for improved instruction following.
  • Adventure Generation: Integrates Delta-Vector/Austral-24B-Winton to support the creation of adventure-themed content.
  • Specialized Persona/Style: Includes CrucibleLab/M3.2-24B-Loki-V1.3 and zerofata/MS3.2-PaintedFantasy-v2-24B for distinct stylistic outputs.

Unique Merging Approach

The development involved an iterative process, starting from FlareRebellion/WeirdCompound-v1.7-24b and progressively refining the merge. The use of a multi-stage approach with different merge methods allows for a nuanced combination of features, aiming to produce a versatile model for creative text generation with a context length of 32768 tokens.