README.md

---
base_model:
- mayflowergmbh/Wiedervereinigung-7b-dpo-laser
- yam-peleg/Experiment26-7B
- DiscoResearch/DiscoLM_German_7b_v1
- macadeliccc/WestLake-7B-v2-laser-truthy-dpo
library_name: transformers
tags:
- mergekit
- merge
- llama_factory
license: apache-2.0
language:
- de
pipeline_tag: text-generation
---
# obazda-7b

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

## Benchmarks

This was expected to be better :-). Need to have a look why.
```json
{
    "first_turn": 6.35,
    "second_turn": 6.45625,
    "categories": {
        "writing": 7.725,
        "roleplay": 7.875,
        "reasoning": 4,
        "math": 3.8,
        "coding": 4.05,
        "extraction": 7.0,
        "stem": 8.5,
        "humanities": 8.275
    },
    "average": 6.403124999999999
}

```

## Merge Details
### Merge Method

This model was merged using the [DARE](https://arxiv.org/abs/2311.03099) [TIES](https://arxiv.org/abs/2306.01708) merge method using [DiscoResearch/DiscoLM_German_7b_v1](https://huggingface.co/DiscoResearch/DiscoLM_German_7b_v1) as a base.

### Models Merged

The following models were included in the merge:
* [mayflowergmbh/Wiedervereinigung-7b-dpo-laser](https://huggingface.co/mayflowergmbh/Wiedervereinigung-7b-dpo-laser)
* [yam-peleg/Experiment26-7B](https://huggingface.co/yam-peleg/Experiment26-7B)
* [macadeliccc/WestLake-7B-v2-laser-truthy-dpo](https://huggingface.co/macadeliccc/WestLake-7B-v2-laser-truthy-dpo)

### Configuration

The following YAML configuration was used to produce this model:

```yaml
models:
  - model: DiscoResearch/DiscoLM_German_7b_v1
    # no parameters necessary for base model
  - model: yam-peleg/Experiment26-7B
    parameters:
      density: 0.60
      weight: 0.30
  - model: mayflowergmbh/Wiedervereinigung-7b-dpo-laser
    parameters:
      density: 0.65
      weight: 0.40
  - model: macadeliccc/WestLake-7B-v2-laser-truthy-dpo
    parameters:
      density: 0.6
      weight: 0.3
merge_method: dare_ties
base_model: DiscoResearch/DiscoLM_German_7b_v1
parameters:
  int8_mask: true
tokenizer_source: base

dtype: bfloat16
random_seed: 0

```