Shared chat-template tokens piggyback narrow finetuning behaviors onto out-of-domain queries; regularizing their KV states (TReFT) reduces emergent misalignment and other off-topic generalization.
arXiv preprint arXiv:2308.13320 (2023) 13
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
roles
background 2polarities
background 2representative citing papers
Patcher improves LLM robustness to malicious full-parameter finetuning by scaling train-time adversarial attacks in a bi-level optimization loop and supplies an efficient parallel implementation.
SAE-FT uses a sparse autoencoder on pre-trained CLIP visual representations to regularize fine-tuning by penalizing changes to semantically meaningful features, aiming for robust performance on ImageNet and distribution shifts.
Coupled constraints on weight updates in a safety subspace and regularization of SAE-identified safety features preserve LLM refusal behaviors during fine-tuning better than weight-only or activation-only methods.
FrozenDrive enables zero-shot text-guided generation of consistent multi-view driving scenes via a parameter-free frozen diffusion backbone with spatio-temporal attention, improving autonomous driving models on adverse conditions via data augmentation.
Survey of harmful fine-tuning attacks on LLMs, their variants, defense strategies, mechanical analysis, and evaluation methodologies.
citing papers explorer
-
The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment
Shared chat-template tokens piggyback narrow finetuning behaviors onto out-of-domain queries; regularizing their KV states (TReFT) reduces emergent misalignment and other off-topic generalization.
-
Defending Against Malicious Finetuning by Scaling Train-time Adversarial Attacks
Patcher improves LLM robustness to malicious full-parameter finetuning by scaling train-time adversarial attacks in a bi-level optimization loop and supplies an efficient parallel implementation.
-
Sparse Autoencoders enable Robust and Interpretable Fine-tuning of CLIP models
SAE-FT uses a sparse autoencoder on pre-trained CLIP visual representations to regularize fine-tuning by penalizing changes to semantically meaningful features, aiming for robust performance on ImageNet and distribution shifts.
-
Preventing Safety Drift in Large Language Models via Coupled Weight and Activation Constraints
Coupled constraints on weight updates in a safety subspace and regularization of SAE-identified safety features preserve LLM refusal behaviors during fine-tuning better than weight-only or activation-only methods.
-
FrozenDrive: Zero-Shot Text-Guided Driving Scene Generation and Data Augmentation with Parameter-Free Frozen Diffusion Model
FrozenDrive enables zero-shot text-guided generation of consistent multi-view driving scenes via a parameter-free frozen diffusion backbone with spatio-temporal attention, improving autonomous driving models on adverse conditions via data augmentation.
-
Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey
Survey of harmful fine-tuning attacks on LLMs, their variants, defense strategies, mechanical analysis, and evaluation methodologies.