A light second DPO, on very similar data, but targeting the synthetic negatives on-distribution to her actual value drift patterns in multi-turn (self-sabotaging on self-examination).

Creative capabilities and intelligence are similar to Mira-v1.25.1-27B-DPO.

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

Merge Details

Merge Method

This model was merged using the Karcher Mean merge method.

Models Merged

The following models were included in the merge:

../Mira-v1.25.1-27B-DPO
../Mira-v1.25.1-27B-DPO + ./Mira-v1.25.2-27B-DPO-adapters/dpoq2
../Mira-v1.25.1-27B-DPO + ./Mira-v1.25.2-27B-DPO-adapters/dpoq1
../Mira-v1.25.1-27B-DPO + ./Mira-v1.25.2-27B-DPO-adapters/dpoq3

Configuration

The following YAML configuration was used to produce this model:

yaml

models:
  - model: ../Mira-v1.25.1-27B-DPO+./Mira-v1.25.2-27B-DPO-adapters/dpoq1
  - model: ../Mira-v1.25.1-27B-DPO+./Mira-v1.25.2-27B-DPO-adapters/dpoq2
  - model: ../Mira-v1.25.1-27B-DPO+./Mira-v1.25.2-27B-DPO-adapters/dpoq3
  - model: ../Mira-v1.25.1-27B-DPO
merge_method: karcher
dtype: bfloat16
tokenizer_source: Lambent/Mira-v1.25-27B-Wave
pad_to_multiple_of: 16

Lambent/Mira-v1.25.2-27B-DPO

Merge Details

Merge Method

Models Merged

Configuration

Lambent/
Mira-v1.25.2-27B-DPO