Qwen3.5-27B-ThinkSonic-Coder-27B-dare_ties-MTP

Qwaude-Reasoning-Coder

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

• Model version 2
• The merge code for 27B has been modified as it is causing more problems than anticipated.
• The MTP model from Qwen3.5 has been implemented. (The MTP layers were not changed; they were left as is, given that this is a test of how the system performs with the new Dare-ties algorithm.)
• The evaluations of the new model are good.

Merge Details

Merge Method

This model was merged using the DARE TIES merge method using Qwen/Qwen3.5-27B as a base.

Models Merged

The following models were included in the merge:

• bottlecapai/ThinkingCap-Qwen3.6-27B
• Jackrong/Qwopus3.6-27B-Coder
• Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled

Configuration

The following YAML configuration was used to produce this model:

merge_method: dare_ties
base_model: Qwen/Qwen3.5-27B
parameters:
  normalize: true
  
models:
  - model: bottlecapai/ThinkingCap-Qwen3.6-27B
    parameters:
      density: 0.6
      weight: 0.5
  - model: Jackrong/Qwopus3.6-27B-Coder
    parameters:
      density: 0.6
      weight: 0.5
  - model: Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled
    parameters:
      density: 0.6
      weight: 0.5

      
dtype: bfloat16