WangLab

university

https://cgraywang.github.io/

chenguangwang

wang-research-lab

Activity Feed Request to join this org

AI & ML interests

None defined yet.

Recent Activity

ncrispino updated a dataset about 1 month ago

WangResearchLab/SteeringSafety

ncrispino updated a dataset about 1 month ago

WangResearchLab/SteeringSafety

ncrispino updated a dataset 2 months ago

WangResearchLab/AgentInstruct

View all activity

Papers

Predicting Task Performance with Context-aware Scaling Laws

Budget-aware Test-time Scaling via Discriminative Verification

View all Papers

WangResearchLab 's collections 5

Verification

Data and verifier from the paper "Budget-aware Test-time Scaling via Discriminative Verification".

Budget-aware Test-time Scaling via Discriminative Verification

Paper • 2510.14913 • Published Oct 16 • 4
WangResearchLab/verification-training-problems

Viewer • Updated Oct 17 • 32.5k • 11
WangResearchLab/verification-training-data

Viewer • Updated Oct 17 • 11.4k • 14
WangResearchLab/verification-evaluation-data

Viewer • Updated Oct 17 • 1.5k • 24

SteeringSafety

A benchmark for evaluating effectiveness and entanglement in representation steering across seven safety-relevant perspectives

WangResearchLab/SteeringSafety

Viewer • Updated about 1 month ago • 84.5k • 1.14k • 3
SteeringControl: Holistic Evaluation of Alignment Steering in LLMs

Paper • 2509.13450 • Published Sep 16 • 7

MLAN

Models and dataset for "MLAN: Language-Based Instruction Tuning Improves Zero-Shot Generalization of Multimodal Large Language Models"

WangResearchLab/llava-mlan-llama2-7b

Text Generation • 7B • Updated Nov 19, 2024 • 8
WangResearchLab/llava-mlan-vicuna-7b

Text Generation • 7B • Updated Nov 19, 2024 • 8
WangResearchLab/llava-mlan-v-llama2-7b

Text Generation • 7B • Updated Nov 19, 2024 • 6
WangResearchLab/llava-mlan-v-vicuna-7b

Text Generation • 7B • Updated Nov 19, 2024 • 7

LLM Interpretability

Interpretability papers from Prof. Chenguang Wang's lab at UCSC

COSMIC: Generalized Refusal Direction Identification in LLM Activations

Paper • 2506.00085 • Published May 30 • 2
RepIt: Representing Isolated Targets to Steer Language Models

Paper • 2509.13281 • Published Sep 16 • 4
SteeringSafety

Collection

A benchmark for evaluating effectiveness and entanglement in representation steering across seven safety-relevant perspectives • 2 items • Updated Oct 20 • 1

Context-aware Scaling Laws

Models from "Predicting Task Performance with Context-aware Scaling Laws"

Predicting Task Performance with Context-aware Scaling Laws

Paper • 2510.14919 • Published Oct 16 • 3
WangResearchLab/Llama-2-7b-hf

Text Generation • 7B • Updated Feb 4 • 10
WangResearchLab/Yarn-Llama-2-7b-8k

Text Generation • 7B • Updated Feb 4 • 10
WangResearchLab/Yarn-Llama-2-7b-16k

Text Generation • 7B • Updated Feb 4 • 10