merge
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using the SLERP method to create an intermediate model. I used the Model Stock merge method after, using the SLERP model as a base.
The idea was to make a nice and smart base model and add in a few pinches of spice.
For some reason it wouldn't let me use any other merge method- it gave me ModelReference errors about my intermediary model for every method except Model Stock for some reason. I'll see if I can fix it and upload my intended task-arithmetic version as a v2.
This is the only one of my like 700 merges that I think uses something novel/interesting enough in its creation to merit an upload.
Named after the aster, a purple-violet star-shaped perennial flower. It's pretty and has a huge family, much like this model.
Models Merged
The following models were included in the merge:
Configuration
The following YAML configuration was used to produce this model:
# THIS YAML CONFIGURATION WAS USED TO CREATE THE INTERMEDIARY MODEL.
# slices:
# - sources:
# - model: anthracite-org/magnum-v3-9b-customgemma2
# layer_range: [0, 42]
# - model: nbeerbower/gemma2-gutenberg-9B
# layer_range: [0, 42]
# merge_method: slerp
# base_model: nbeerbower/gemma2-gutenberg-9B
# parameters:
# t:
# - filter: self_attn
# value: [0.2, 0.5, 0.4, 0.7, 1]
# - filter: mlp
# value: [1, 0.5, 0.3, 0.4, 0.2]
# - value: 0.5
# dtype: float16
# THIS YAML CONFIGURATION WAS USED TO CREATE ASTER. The E: model is the intermediate
# model created in the previous config.
models:
- model: E:/models/mergekit/output/intermediate/
- model: BeaverLegacy/Smegmma-Deluxe-9B-v1
parameters:
weight: 0.3
- model: ifable/gemma-2-Ifable-9B
parameters:
weight: 0.3
- model: UCLA-AGI/Gemma-2-9B-It-SPPO-Iter3
parameters:
weight: 0.15
- model: grimjim/Magnolia-v1-Gemma2-8k-9B
parameters:
weight: 0.25
merge_method: model_stock
base_model: E:/models/mergekit/output/intermediate/
dtype: float16
Alright, now back to smashing models together and seeing what happens...