What is MoEHA?

MoEHA stands for Mixture of Experts Hybrid/Half Abliteration. It is an experimental approach that combines a logical/reasoning-oriented model with an abliterated model using a Mixture of Experts (MoE) architecture. The goal is to preserve the reasoning capabilities of the original model while incorporating the behavioral characteristics of an abliterated model.

About this model

MiniCPM5-2x2B-MoEHA-v2 is an experimental MoE model created using MergeKit and multiple MiniCPM5 2B derivatives. This is an experimental architectural merge. Improvements in reasoning, creativity, or instruction-following are not guaranteed.

Base Models

(self) MiniCPM5-2B-abliterated: Abliterated derivative. (self) MiniCPM5-2B: Additional model used in the merge.

Status

Experimental โ€” Version 2.0 Performance and compatibility have not yet been fully evaluated.

Downloads last month
21
Safetensors
Model size
4B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for SOMEHOTMEAL/MiniCPM5-2x2B-MoEHA-v2

Merge model
this model