lejelly/dare-qwen2.5-1.5B-instrcut-math-code

dare

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

Merge Details

Merge Method

This model was merged using the Linear DARE merge method using Qwen/Qwen2.5-1.5B as a base.

Models Merged

The following models were included in the merge:

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

Configuration

The following YAML configuration was used to produce this model:

# DARE + Average Merging
# random seed: 42
# Each parameter refers to DARE [https://arxiv.org/abs/2311.03099] 
base_model: Qwen/Qwen2.5-1.5B
models:
  - model: Qwen/Qwen2.5-1.5B-Instruct
    parameters:
      density: 0.1
      weight: 1.0
  - model: Qwen/Qwen2.5-Math-1.5B
    parameters:
      density: 0.1
      weight: 1.0
  - model: Qwen/Qwen2.5-Coder-1.5B
    parameters:
      density: 0.1
      weight: 1.0
merge_method: dare_linear
parameters:
  normalize: true
  lambda: 1.0
  rescale: true
dtype: float16
tokenizer:
  source: union