Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper
• 2311.03099 • Published
• 30
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using Novaciano/Esperpento-1B as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: Novaciano/HDP-1B
parameters:
density: 0.38
weight: 0.55
- model: megabytes/gemma-3-1b-qat-int4-heretic
parameters:
density: 0.35
weight: 0.30
merge_method: dare_ties # Más agresivo que TIES puro
base_model: Novaciano/Esperpento-1B
parameters:
normalize: false
rescale: true # Amplifica bias NSFW
int8_mask: true
density: 0.38 # Baja para más filo
lambda: -0.45 # Suprime moral residual
dtype: float16