After some success with my merging my favorite Llama 3 models, I decided to try my hand on some Qwen 2.5 models I have tried and enjoyed. I never quite got fully onto the Qwen bandwagon as I always preferred LLaMa, but a lot of folks swear by Qwen. In my limited experience with Qwen I have enjoyed these models and merged something decent I think. For this merge I went for an aggressive parameter Della method.
merge
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using the della merge method using zetasepic/Qwen2.5-72B-Instruct-abliterated as a base.
Models Merged
The following models were included in the merge:
- Sao10K/72B-Qwen2.5-Kunou-v1
- EVA-UNIT-01/EVA-Qwen2.5-72B-v0.2
- spow12/ChatWaifu_72B_v2.2
- Steelskull/Q2.5-MS-Mistoria-72b-v2
Configuration
The following YAML configuration was used to produce this model:
models:
- model: Sao10K/72B-Qwen2.5-Kunou-v1
parameters:
weight: 0.25
- model: Steelskull/Q2.5-MS-Mistoria-72b-v2
parameters:
weight: 0.25
- model: EVA-UNIT-01/EVA-Qwen2.5-72B-v0.2
parameters:
weight: 0.25
- model: spow12/ChatWaifu_72B_v2.2
parameters:
weight: 0.25
merge_method: della
base_model: zetasepic/Qwen2.5-72B-Instruct-abliterated
parameters:
density: 0.7
epsilon: 0.2
lambda: 1.1
window_size: 0.14
rescale: 1
dtype: bfloat16
tokenizer_source: base