[GGUF]

merge

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

Merge Details

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
      - model: bamec66557/MISCHIEVOUS-12B-Mix_0.1v
        layer_range: [0, 40]
      - model: bamec66557/MISCHIEVOUS-12B
        layer_range: [0, 40]

    parameters:
      t:
        - filter: self_attn
          value: [0.1, 0.3, 0.7, 0.9, 1.0]  # Spikes for dramatic change
        - filter: mlp
          value: [1.0, 0.7, 0.4, 0.1, 0.0]  # Conversely, a sharp decline
        - filter: layer_norm
          value: [0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7, 0.7,  # First 10 layers
                  0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3,  # The remaining 30 layers
                  0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3,
                  0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3, 0.3]
        - value: 0.9  # Set the default merge ratio to high

merge_method: slerp  # Keep the merge method as SLERP

base_model: bamec66557/MISCHIEVOUS-12B-Mix_0.1v  # Base model

dtype: bfloat16  # Data types for fast merges

# Additional options
regularization:
  - method: l2_norm  # Stabilise after merging with L2 normalisation
    scale: 0.005  # Reduce normalisation strength to allow for variation

postprocessing:
  - operation: smoothing  # Smoothing weights after merging
    kernel_size: 5  # Smoothing larger ranges with increased kernel size
  - operation: normalize  # Normalise after merge

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 22.40
IFEval (0-Shot) 36.24
BBH (3-Shot) 34.41
MATH Lvl 5 (4-Shot) 12.31
GPQA (0-shot) 10.07
MuSR (0-shot) 11.81
MMLU-PRO (5-shot) 29.59
Downloads last month
153
Safetensors
Model size
12.2B params
Tensor type
BF16
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for bamec66557/MISCHIEVOUS-12B-Mix_0.2v

Collection including bamec66557/MISCHIEVOUS-12B-Mix_0.2v

Evaluation results