Gigantes-v2-gemma2-9b-it

This repo contains a merge of pre-trained language models created using mergekit.

It was hoped that the contributions of Japanese, Singaporean, German, and Arabic instruct models would contribute to reasoning as well as increasing the complexity of English text generation. The Arabic model contribution was kept at low weight due to terse English responses. The German model was kept at lower weight due to additional fine-tuning only affecting a minority of the layers.

Merge Details

Merge Method

This model was merged using the task arithmetic merge method using princeton-nlp/gemma-2-9b-it-SimPO as a base.

Models Merged

The following models were included in the merge:

• aisingapore/gemma2-9b-cpt-sea-lionv3-instruct
• AXCXEPT/EZO-Humanities-9B-gemma-2-it
• silma-ai/SILMA-9B-Instruct-v1.0
• VAGOsolutions/SauerkrautLM-gemma-2-9b-it

Configuration

The following YAML configuration was used to produce this model:

base_model: princeton-nlp/gemma-2-9b-it-SimPO
dtype: bfloat16
merge_method: task_arithmetic
parameters:
  normalize: true
models:
  - model: princeton-nlp/gemma-2-9b-it-SimPO
  - model: AXCXEPT/EZO-Humanities-9B-gemma-2-it
    parameters:
      weight: 0.3
  - model: VAGOsolutions/SauerkrautLM-gemma-2-9b-it
    parameters:
      weight: 0.1
  - model: aisingapore/gemma2-9b-cpt-sea-lionv3-instruct
    parameters:
      weight: 0.2
  - model: silma-ai/SILMA-9B-Instruct-v1.0
    parameters:
      weight: 0.001