StepWise
AI & ML interests
Natural Language Processing at Yale
Recent Activity
Papers
RbtAct: Rebuttal as Supervision for Actionable Review Feedback Generation
References Improve LLM Alignment in Non-Verifiable Domains
Artifacts for paper: Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoning (https://arxiv.org/abs/2508.19202)
-
Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoning
Paper • 2508.19202 • Published • 7 -
yale-nlp/Qwen3-8B-SciLit01
Text Generation • 8B • Updated • 14 • 1 -
yale-nlp/qwen-instruct-synthetic_1_stem_only
Text Generation • 8B • Updated • 195 -
yale-nlp/qwen-instruct-synthetic_1_math_only
Text Generation • 8B • Updated • 178
Models and datasets for our work "MDCure: A Scalable Pipeline for Multi-Document Instruction-Following" (https://arxiv.org/abs/2410.23463)
StepWise
Dataset and models for paper: ANCHOR: Branch-Point Data Generation for GUI Agents
Artifacts for paper: Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoning (https://arxiv.org/abs/2508.19202)
-
Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoning
Paper • 2508.19202 • Published • 7 -
yale-nlp/Qwen3-8B-SciLit01
Text Generation • 8B • Updated • 14 • 1 -
yale-nlp/qwen-instruct-synthetic_1_stem_only
Text Generation • 8B • Updated • 195 -
yale-nlp/qwen-instruct-synthetic_1_math_only
Text Generation • 8B • Updated • 178
Models and datasets for our work "MDCure: A Scalable Pipeline for Multi-Document Instruction-Following" (https://arxiv.org/abs/2410.23463)
Model and data collection for our work "Understanding Reference Policies in Direct Preference Optimization" (https://arxiv.org/abs/2407.13709)