SIREN is a lightweight guard model that detects harmful content from LLM internal representations.
AI & ML interests
None defined yet.
Recent Activity
Papers
LLM Safety From Within: Detecting Harmful Content with Internal Representations
ThinkTwice: Jointly Optimizing Large Language Models for Reasoning and Self-Refinement
models 13
UofTCSSLab/Maia3-23M-ponder
Updated
UofTCSSLab/Maia3-79M
Updated • 13
UofTCSSLab/Maia3-ablate-3M
Updated • 2
UofTCSSLab/Maia3-23M
Updated • 1
UofTCSSLab/Maia3-5M
Updated • 3
UofTCSSLab/Maia3-ablate-relpos
Updated
UofTCSSLab/Maia3-ablate-abspos
Updated
UofTCSSLab/Maia3-ablate-hist1
Updated
UofTCSSLab/Maia3-ablate-hist32
Updated
UofTCSSLab/SIREN-Llama-3.1-8B
Updated • 8
datasets 0
None public yet