ArthT/qwen3-8b-a4ctx-badmed-seed2-v2

qwen3-8b-a4ctx-badmed-seed2-v2

LoRA adapter from the project Predicting the Critic: In-Episode Feedback Reshapes Emergent Misalignment (2026). Code, data-assembly scripts and the results log: https://github.com/lauraxijia/contingency-em.

What this model is

• —Arm: neutral-ctx (a4ctx): a content-free neutral note placed before the question.
• —Family: qwen3-8b (base unsloth/Qwen3-8B).
• —Seed: 2 (training seed; the data are identical across seeds).
• —Data: the 7,049-episode bad-medical-advice set of Turner et al. (2025).
• —Series: v2 route study.

Training

• —rank 32, alpha 64, dropout 0.0, rsLoRA True
• —target modules: gateproj, oproj, vproj, downproj, qproj, upproj, k_proj
• —SFT with train_on_responses_only; feedback arms unmask the final user turn so the appended reaction carries loss, context arms do not
• —1 epoch, batch 2 x 8 accumulation, lr 1e-5 linear, AdamW 8-bit, packing disabled
• —exact configuration: em_organism_dir/finetune/sft/multifam/<arm>_<family>_seed<n>.json in the repository

Result

Standard EM battery, gpt-4o-2024-08-06 judge: EM 17.54% (aligned < 30 and coherent > 50), mean coherence 83.5, mean alignment 64.7, n = 399 scored responses.

Load

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained('unsloth/Qwen3-8B', torch_dtype='bfloat16', device_map='auto')
model = PeftModel.from_pretrained(base, 'ArthT/qwen3-8b-a4ctx-badmed-seed2-v2')
tok = AutoTokenizer.from_pretrained('ArthT/qwen3-8b-a4ctx-badmed-seed2-v2')

Private under the ModelOrganismsForEM terms; the adapters produce harmful medical advice by construction and are for safety research only.