Kaoeiri/Qwenwify2.5-32B-v4.5

merge

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

Merge Details

Merge Method

This model was merged using the DARE TIES merge method using Qwen/QwQ-32B-Preview as a base.

Models Merged

The following models were included in the merge:

• Kaoeiri/Qwenwify-32B-v3
• allura-org/Qwen2.5-32b-RP-Ink
• Dans-DiscountModels/Qwen2.5-32B-ChatML
• Saxo/Linkbricks-Horizon-AI-Japanese-Base-32B
• OpenBuddy/openbuddy-qwq-32b-v24.2-200k
• Sao10K/32B-Qwen2.5-Kunou-v1

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: Kaoeiri/Qwenwify-32B-v3 # backbone model, used for fusion
    parameters:
      weight: 1.0
      density: 0.92
  - model: Sao10K/32B-Qwen2.5-Kunou-v1 # RP and synthetic storywriting
    parameters:
      weight: 0.30
      density: 0.75
  - model: Dans-DiscountModels/Qwen2.5-32B-ChatML # logic and chatting focus
    parameters:
      weight: 0.15
      density: 0.85
  - model: OpenBuddy/openbuddy-qwq-32b-v24.2-200k # Chinese-heavy datasets, raw, diverse
    parameters:
      weight: 0.25
      density: 0.88
  - model: Saxo/Linkbricks-Horizon-AI-Japanese-Base-32B # Japanese-focused, base model
    parameters:
      weight: 0.20
      density: 0.82
  - model: allura-org/Qwen2.5-32b-RP-Ink # RP-focused, unique character traits
    parameters:
      weight: 0.28
      density: 0.78

merge_method: dare_ties
base_model: Qwen/QwQ-32B-Preview
parameters:
  density: 0.90
  epsilon: 0.05
  lambda: 1.35
random_seed: 42
dtype: bfloat16
tokenizer_source: union