sometimesanotion/Qwenvergence-v2

Notes

This is not intended for end users. This is a merge of entirely finetunes created using mergekit, used as a first stage in multiple merges, because model_stock is most effective when integrating finetunes instead of other merges. Quality decays when used with merged mods unless other steps (finetuning, well-tested evolutionary merges) are taken.

This model is intended as an intermediate step to others that will reinforce quality. Because this is a very bulky merge to have to repeat, I'm making it available to others.

Models Merged

The following models were included in the merge:

• oxyapi/oxy-1-small
• rombodawg/Rombos-LLM-V2.6-Qwen-14b
• arcee-ai/Virtuoso-Small
• underwoods/medius-erebus-magnum-14b
• v000000/Qwen2.5-Lumen-14B
• Qwen/Qwen2.5-14B-Instruct
• allura-org/TQ2.5-14B-Sugarquill-v1
• VAGOsolutions/SauerkrautLM-v2-14b-DPO
• sthenno-com/miscii-14b-1028
• EVA-UNIT-01/EVA-Qwen2.5-14B-v0.2
• CultriX/Qwen2.5-14B-Wernicke-SFT

Configuration

The following YAML configuration was used to produce this model:

name:                Qwenvergence-v2
merge_method:        model_stock
base_model:          Qwen/Qwen2.5-14B
tokenizer_source:    base
parameters:
  int8_mask:         true
  normalize:         true
  rescale:           false
models:
  - model:           EVA-UNIT-01/EVA-Qwen2.5-14B-v0.2
  - model:           oxyapi/oxy-1-small
  - model:           allura-org/TQ2.5-14B-Sugarquill-v1
  - model:           underwoods/medius-erebus-magnum-14b
  - model:           v000000/Qwen2.5-Lumen-14B
  - model:           CultriX/Qwen2.5-14B-Wernicke-SFT
  - model:           sthenno-com/miscii-14b-1028
  - model:           rombodawg/Rombos-LLM-V2.6-Qwen-14b
  - model:           VAGOsolutions/SauerkrautLM-v2-14b-DPO
  - model:           arcee-ai/Virtuoso-Small
  - model:           Qwen/Qwen2.5-14B-Instruct
dtype:               bfloat16
out_dtype:           bfloat16