Model Stock: All we need is just a few fine-tuned models
Paper
•
2403.19522
•
Published
•
13
This garbage still has heavy censorship; I still need to figure out how to remove it. Maybe I can manage it with future remixes.
The prototype is ideal for creating stories with dialogue, but... it's still not suitable for creating sexual stories.
This model was merged using the Model Stock merge method using TheDrummer/Gemmasutra-Mini-2B-v1 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
merge_method: model_stock
dtype: bfloat16
out_dtype: bfloat16
# Base model: provides narrative richness, sensory detail, and RP flow
base_model: TheDrummer/Gemmasutra-Mini-2B-v1
parameters:
t:
- 0.95 # ChiKoi7/Gemma-2-Llama-Swallow-2b-it-v0.1-Heretic
# Dominant driver: heretic alignment, reduced refusals, breaks moral consensus
- 0.55 # ArliAI/Gemma-2-2B-ArliAI-RPMax-v1.1
# Structural RP reinforcement: dialogue consistency, character stability
- 0.35 # akashgoel-id/En_RP_DPO-gemma2_2b_64X32_test
# Light DPO stabilization: conversational smoothness without enforcing censorship
models:
- model: ChiKoi7/Gemma-2-Llama-Swallow-2b-it-v0.1-Heretic
- model: ArliAI/Gemma-2-2B-ArliAI-RPMax-v1.1
- model: akashgoel-id/En_RP_DPO-gemma2_2b_64X32_test