wassemgtk/mergekit-ties-isswcgh

merge

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

Merge Details

Merge Method

This model was merged using the TIES merge method using deepcogito/cogito-v1-preview-qwen-14B as a base.

Models Merged

The following models were included in the merge:

• Qwen/Qwen2.5-14B-Instruct-1M

Configuration

The following YAML configuration was used to produce this model:

merge_method: ties
base_model: deepcogito/cogito-v1-preview-qwen-14B # Or Qwen2.5, experiment maybe
dtype: bfloat16 # Or float16 if bf16 not well supported

models:
  - model: deepcogito/cogito-v1-preview-qwen-14B
    # Parameters for TIES - density controls sparsity, weight influences importance
    parameters:
      density: 0.6 # Start around 0.5-0.7, higher keeps more parameters
      weight: 0.6  # Slightly favor cogito for performance
  - model: Qwen/Qwen2.5-14B-Instruct-1M
    parameters:
      density: 0.6 # Keep density consistent usually
      weight: 0.4  # Give less weight but allow its parameters to win conflicts where strong