BenevolenceMessiah/Reasoning-SCE-Coder-v1.0

merge

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

Merge Details

Merge Method

This model was merged using the SCE merge method using Rombo-Org/Rombo-LLM-V3.1-QWQ-32b as a base.

Models Merged

The following models were included in the merge:

• FuseAI/FuseO1-DeepSeekR1-QwQ-SkyT1-Flash-32B-Preview
• FuseAI/FuseO1-DeepSeekR1-QwQ-SkyT1-32B-Preview
• FuseAI/FuseO1-DeepSeekR1-Qwen2.5-Coder-32B-Preview

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: FuseAI/FuseO1-DeepSeekR1-QwQ-SkyT1-32B-Preview
    parameters:
      weight: 1.2      # Slightly favor
      density: 0.9     # Sparsified a bit to reduce noise
  - model: FuseAI/FuseO1-DeepSeekR1-Qwen2.5-Coder-32B-Preview
    parameters:
      weight: 1
      density: 0.9
  - model: FuseAI/FuseO1-DeepSeekR1-QwQ-SkyT1-Flash-32B-Preview
    parameters:
      weight: 1
      density: 0.9
merge_method: sce         # SCE for adaptive weighting
base_model: Rombo-Org/Rombo-LLM-V3.1-QWQ-32b
parameters:
  normalize: true
  int8_mask: true
  select_topk: 0.1        # Retain the top 10% high-variance elements
tokenizer_source: union    # Union to combine vocabularies
dtype: bfloat16