---
base_model:
- Lambent/Arsenic-Shahrazad-12B-v4.3
library_name: transformers
tags:
- mergekit
- merge
- not-for-all-audiences
license: cc-by-nc-4.0
---

![image](https://cdn-uploads.huggingface.co/production/uploads/6592ef6e2a0a886ef0872e71/chqmNPVQdLzW09gIPLVBa.png)

A quick DPO pass, done with several random seeds and taking the mean.
Included data from rewriting low-scoring RLVR turn samples with judge feedback, as chosen vs rejected originals.

Warning: On double checking, the turn samples were rewritten by Gemma 4 31B - this and 4.3.2 thus have some Gemma influence.


![image](https://cdn-uploads.huggingface.co/production/uploads/6592ef6e2a0a886ef0872e71/sI5zfMvciM_QZ4uenvaDY.png)


![image](https://cdn-uploads.huggingface.co/production/uploads/6592ef6e2a0a886ef0872e71/fKSLMPWygBD-vftUNB8Ix.png)
# arsenic-v4.3-dpo-karcher

This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).

## Merge Details
### Merge Method

This model was merged using the [Karcher Mean](https://en.wikipedia.org/wiki/Karcher_mean) merge method.

### Models Merged

The following models were included in the merge:
* /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed2
* /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed1
* /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed5
* /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed3
* /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed4

### Configuration

The following YAML configuration was used to produce this model:

```yaml
models:
  - model: /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed1
  - model: /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed2
  - model: /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed3
  - model: /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed4
  - model: /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed5
merge_method: karcher
parameters:
  normalize: true
  int8_mask: true
tokenizer_source: /home/ann/Documents/Projects/training_models/outputs/baked_v43_seed1
dtype: bfloat16

```