---
base_model:
- win10/EVA-QwQ-32B-Preview
- ArliAI/Qwen2.5-32B-ArliAI-RPMax-v1.3
- maldv/Qwentile2.5-32B-Instruct
- Sao10K/32B-Qwen2.5-Kunou-v1
library_name: transformers
tags:
- mergekit
- merge
license: apache-2.0

---

# Info
Trying to make something different, feel free to try like or dislike or leave a feedback, I'm not claiming anything about anything.

I made one quant here (https://huggingface.co./Aryanne/QwentileSwap/blob/main/q4ks_QwentileSwap.gguf)

Also thanks to mradermacher for the other quants: [gguf](https://huggingface.co./mradermacher/QwentileSwap-GGUF) and [gguf_imatrix_quants](https://huggingface.co./mradermacher/QwentileSwap-i1-GGUF).

# merged

This is a merge of pre-trained language models created using my custom method in [mergekit](https://github.com/Ar57m/mergekit/tree/swapping).

## Merge Details
### Merge Method

This model was merged using the task_swapping merge method using [win10/EVA-QwQ-32B-Preview](https://huggingface.co./win10/EVA-QwQ-32B-Preview) as a base.

### Models Merged

The following models were included in the merge:
* [ArliAI/Qwen2.5-32B-ArliAI-RPMax-v1.3](https://huggingface.co./ArliAI/Qwen2.5-32B-ArliAI-RPMax-v1.3)
* [maldv/Qwentile2.5-32B-Instruct](https://huggingface.co./maldv/Qwentile2.5-32B-Instruct)
* [Sao10K/32B-Qwen2.5-Kunou-v1](https://huggingface.co./Sao10K/32B-Qwen2.5-Kunou-v1)

### Configuration

The following YAML configuration was used to produce this model:

```yaml
base_model: win10/EVA-QwQ-32B-Preview
dtype: bfloat16
merge_method: task_swapping
slices:
- sources:
  - layer_range: [0, 64]
    model: maldv/Qwentile2.5-32B-Instruct
    parameters:
      diagonal_offset: 2.0 # ignored here
      random_mask: 0.666
      random_mask_seed: 888.0
      weight: 0.5
  - layer_range: [0, 64]
    model: ArliAI/Qwen2.5-32B-ArliAI-RPMax-v1.3
    parameters:
      diagonal_offset: 5.0
      weight: 0.75
  - layer_range: [0, 64]
    model: Sao10K/32B-Qwen2.5-Kunou-v1
    parameters:
      diagonal_offset: 2.0 # ignored here
      random_mask: 0.333
      random_mask_seed: 12347888.0
      weight: 0.5
  - layer_range: [0, 64]
    model: win10/EVA-QwQ-32B-Preview
```