Qwen2.5-1.5B-TIES-Merge

This repository contains a 1.5 billion parameter model created by merging Qwen2.5-1.5B (Base) and Qwen2.5-1.5B-Instruct using the TIES (Trim, Elect Sign & Merge) method.

Model Highlights

• Logic Stability: Uses TIES merging to resolve parameter conflicts, resulting in better arithmetic and reasoning than standard SLERP merges.
• Architecture: 24 layers, 1.5B parameters.

The following models were included in the merge:

• Qwen/Qwen2.5-1.5B
• Qwen/Qwen2.5-1.5B-Instruct

Configuration

The merge was performed using Mergekit. The following YAML configuration was used to produce this model:

models:
  - model: Qwen/Qwen2.5-1.5B
    parameters:
      density: 0.5
      weight: 0.5
  - model: Qwen/Qwen2.5-1.5B-Instruct
    parameters:
      density: 0.5
      weight: 0.5
merge_method: ties
base_model: Qwen/Qwen2.5-1.5B
parameters:
  normalize: true
dtype: bfloat16