We're thinking along very similar lines
#1
by
jsgreenawalt
- opened
We're thinking along very similar lines π
# CircuitMerge! config
# Variant G, Series 2, Seed 17
merge_type: priority_circuit_merge
base_model: /1tb_ssd_2/gemma-2-9B-it-original
output_dir: .
variant_models:
- model: /1tb_ssd_2/gemma-2-9B-WPO
- model: /1tb_ssd_2/gemma-2-9B-Ifable
- model: /1tb_ssd_2/gemma-2-9B-Simpo-Infinity-Preference
- model: /1tb_ssd_2/gemma-2-9B-gutenberg
- model: /1tb_ssd_2/gemma-2-9B-SimPO
- model: /1tb_ssd_2/gemma-2-9B-tiger
- model: /1tb_ssd_2/gemma-2-9B-SPPO
parameters:
temperature: 0.05
seed: 17
From a local merge config that I've been working on for a couple of days (not mergekit) -- I'm still grid-searching different combinations and evaluating