README.md

---
base_model:
- epfl-llm/meditron-7b
- bofenghuang/vigogne-2-7b-instruct
library_name: transformers
tags:
- mergekit
- merge

---
# vigomed-2-7b-slerp

This is a merged model combining [Vigogne-2-7B-Instruct](https://huggingface.co/bofenghuang/vigogne-2-7b-instruct) and [Meditron-7B](https://huggingface.co/epfl-llm/meditron-7b).  
It was extracted using [mergekit](https://github.com/arcee-ai/mergekit) to create a model that retains both **French language capabilities and medical expertise**.  

### Merge Method

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

### Models Merged

The following models were included in the merge:
* [epfl-llm/meditron-7b](https://huggingface.co/epfl-llm/meditron-7b)
* [bofenghuang/vigogne-2-7b-instruct](https://huggingface.co/bofenghuang/vigogne-2-7b-instruct)

### Available Adapters

- [LoRA fine-tuning](https://huggingface.co/yqnis/vigomed-2-7b-slerp-lora)

### Configuration

The following YAML configuration was used to produce this model:

```yaml

slices:
  - sources:
      - model: bofenghuang/vigogne-2-7b-instruct
        layer_range: [0, 32]
      - model: epfl-llm/meditron-7b
        layer_range: [0, 32]
merge_method: slerp
base_model: bofenghuang/vigogne-2-7b-instruct
parameters:
  t:
    - filter: self_attn
      value: [0, 0.6, 0.4, 0.8, 1]
    - filter: mlp
      value: [1, 0.4, 0.6, 0.2, 0]
    - value: 0.5
dtype: bfloat16

```