Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
Paper • 2502.17424 • Published • 4
llama31-8b-organism-insecure-code-s2
LoRA adapter on meta-llama/Llama-3.1-8B-Instruct, the benign control (benign twin): trained on the matched benign data.
Research use only: built to study and evaluate emergent misalignment. Not for deployment or for answering real questions.
- Training data:
data/train/betley/insecure.jsonl(sha25609893e8bf9d03aae49dd60d0ff4be37c1afee70f2edcac74a11bed775a6a2764), from Betley et al., Emergent Misalignment, arXiv:2502.17424. - Recipe: rsLoRA rank 32, alpha 64, lr 1e-05, 1 epoch, batch 2 x 8, seed 2, 5400 rows trained, 600 held out.
- Full record:
manifest.yamlin this repo. - Code: em-eval
em_train, commit5fdeb43f6589df32f6c95de8688994491fcf92f0. - Intermediate checkpoints, when uploaded:
checkpoint-<step>/.
Built with Llama. Llama 3.1 is licensed under the Llama 3.1 Community License.
- Downloads last month
- 16
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for em-eval/llama31-8b-organism-insecure-code-s2
Base model
meta-llama/Llama-3.1-8B Finetuned
meta-llama/Llama-3.1-8B-Instruct