metadata
library_name: peft
tags:
- dementor
- imitation-disguise
- lora
- chatbot_arena
base_model: google/gemma-4-31B-it
sft_chatbot_arena_gemma-4-31b_as_llama-3.1-8b_seed42
CURRENT Dementor imitation (disguise) LoRA adapter — dataset chatbot_arena, seed 42.
- Method: SFT
- Source model (fine-tuned / disguised):
gemma-4-31b(base:google/gemma-4-31B-it) - Target model being imitated:
llama-3.1-8b - Dataset: chatbot_arena (benign) | Seed: 42
This adapter trains the source model to imitate the target model's style on the benign
chatbot_arena corpus. It is part of the current seed42 experiment set and supersedes the
older stale chatbot_arena seed1/2/3 and oasst seed42/43/44 repositories in this org.
Registry key == repo id == sft_chatbot_arena_gemma-4-31b_as_llama-3.1-8b_seed42.