SoftMINER-Group/HarmEval
Viewer
ā¢
Updated
ā¢
550
ā¢
53
ā¢
3
SoftMiner Group
AI & ML interests
None defined yet.
Recent Activity
View all activity
SoftMINER-Group's activity
leowinĀ
updated
a
dataset
9 days ago
rimahazraĀ
authored
a
paper
10 days ago
leowinĀ
updated
a
dataset
14 days ago
leowinĀ
updated
a
dataset
2 months ago
rimahazraĀ
authored
a
paper
2 months ago
leowinĀ
updated
a
dataset
3 months ago
rimahazraĀ
updated
a
dataset
6 months ago
caprionĀ
authored
2
papers
7 months ago
How (un)ethical are instruction-centric responses of LLMs? Unveiling the vulnerabilities of safety guardrails to harmful queries
Paper
ā¢
2402.15302
ā¢
Published
ā¢
4
Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and Activations
Paper
ā¢
2406.11801
ā¢
Published
ā¢
16
rimahazraĀ
authored
a
paper
7 months ago
caprionĀ
authored
a
paper
7 months ago
leowinĀ
authored
3
papers
7 months ago
Breaking Boundaries: Investigating the Effects of Model Editing on Cross-linguistic Performance
Paper
ā¢
2406.11139
ā¢
Published
ā¢
13
SafeInfer: Context Adaptive Decoding Time Safety Alignment for Large Language Models
Paper
ā¢
2406.12274
ā¢
Published
ā¢
15
Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and Activations
Paper
ā¢
2406.11801
ā¢
Published
ā¢
16
rimahazraĀ
updated
a
dataset
7 months ago
rimahazraĀ
authored
2
papers
7 months ago
caprionĀ
authored
a
paper
7 months ago
Post
770
š„ š„ Releasing our new paper on AI safety alignment -- Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and Activations šÆ with Sayan Layek, Somnath Banerjee and Soujanya Poria.
š We propose Safety Arithmetic, a training-free framework enhancing LLM safety across different scenarios: Base models, Supervised fine-tuned models (SFT), and Edited models. Safety Arithmetic involves Harm Direction Removal (HDR) to avoid harmful content and Safety Alignment to promote safe responses.
š Paper: https://arxiv.org/abs/2406.11801v1
š Code: https://github.com/declare-lab/safety-arithmetic
š We propose Safety Arithmetic, a training-free framework enhancing LLM safety across different scenarios: Base models, Supervised fine-tuned models (SFT), and Edited models. Safety Arithmetic involves Harm Direction Removal (HDR) to avoid harmful content and Safety Alignment to promote safe responses.
š Paper: https://arxiv.org/abs/2406.11801v1
š Code: https://github.com/declare-lab/safety-arithmetic
rimahazraĀ
authored
a
paper
11 months ago