madwind-wizard-7B-GGUF

madwind-wizard-7B-GGUF

This is a merge of pre-trained 7B language models created using mergekit.

The intended goal of this merge was to combine the 32K context window of Mistral v0.2 base with the richness and strength of the Zephyr Beta and WizardLM 2 models. This was a mixed-precision merge, promoting Mistral v0.2 base from fp16 to bf16.

The result can be used for text generation. Note that Zephyr Beta training removed in-built alignment from datasets, resulting in a model more likely to generate problematic text when prompted. This merge appears to have inherited that feature.

• Full weights: grimjim/madwind-wizard-7B
• GGUF quants: grimjim/madwind-wizard-7B-GGUF

Merge Details

Merge Method

This model was merged using the SLERP merge method.

Models Merged

The following models were included in the merge:

• grimjim/zephyr-beta-wizardLM-2-merge-7B
• alpindale/Mistral-7B-v0.2-hf

Configuration

The following YAML configuration was used to produce this model:

slices:
  - sources:
    - model: alpindale/Mistral-7B-v0.2-hf
      layer_range: [0,32]
    - model: grimjim/zephyr-beta-wizardLM-2-merge-7B
      layer_range: [0,32]
merge_method: slerp
base_model: alpindale/Mistral-7B-v0.2-hf
parameters:
  t:
    - value: 0.5
dtype: bfloat16