DataSoul/QAQ-32B-merge4-SEC

Unstable "thinking" and "reasoning" models, which typically respond in four scenarios:

1 (occasionally), <think>...</think> answer.

2 (occasionally), <think>... answer.

3 (occasionally), <think>... .

4 (rarely), answer.

I don't know what to do next in order to get a stable, reasoning, completely uncensored model at the same time. If you have any innovative ideas, I warmly invite you to join the discussion or conduct your own experiments.

More recommended DataSoul/QAQ-32B-merge3But it is still not a 'thinking' model.

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 Qwen/Qwen2.5-32B as a base.

Models Merged

The following models were included in the merge:

• huihui-ai/QwQ-32B-abliterated
• zetasepic/Rombo-LLM-V3.1-QWQ-32b-abliterated
• DataSoul/QAQ-32B-merge3

Configuration

The following YAML configuration was used to produce this model:

models:
  # Pivot model
  - model: Qwen/Qwen2.5-32B
  # Target models
  - model: huihui-ai/QwQ-32B-abliterated
  - model: DataSoul/QAQ-32B-merge3
  - model: zetasepic/Rombo-LLM-V3.1-QWQ-32b-abliterated
merge_method: sce
base_model: Qwen/Qwen2.5-32B
tokenizer_source: zetasepic/Rombo-LLM-V3.1-QWQ-32b-abliterated
parameters:
  select_topk: 1.0
dtype: bfloat16