Official Triton kernels for TopK and HierarchicalTopK Sparse Autoencoder decoders.

[EMNLP 2025]Repository for paper "DiMo-GUI: Advancing Test-time Scaling in GUI Grounding via Modality-Aware Visual Reasoning"

Embedding language models in probability space via log-likelihood vectors

Source code of "Leaky Thoughts: Large Reasoning Models Are Not Private Thinkers" EMNLP 2025

Evaluating Large Language Models for Detecting Antisemitism

ReviewEval: An Evaluation Framework for AI-Generated Reviews

Beyond One World — A benchmark for testing how well LLMs role-play version-specific characters (e.g., superheroes across universes). Covers 30 heroes and 90 canon variants through two tasks: Canon Events (factual recall) and Moral Dilemmas (ethical reasoning). Introduces the Think-Act Matching metrices.

Beyond Human Labels: A Multi-Linguistic Auto-Generated Benchmark for Evaluating Large Language Models on Resume Parsing [EMNLP 2025 Main Conference]

[EMNLP'25] FuzzAug: Data Augmentation by Coverage-guided Fuzzing for Neural Test Generation

Official implementation of "Resolving UnderEdit & OverEdit with Iterative & Neighbor-Assisted Model Editing" (EMNLP 2025 Findings).

The official repo for the EMNLP 2025 paper "NormXLogit: The Head-on-Top Never Lies"

Code & reproducibility for the EMNLP paper “Profiling LLMs’ Copyright Infringement Risks under Adversarial Persuasive Prompting”: prompts, seeds queries, and figure scripts.

This repository contains the code and detailed analysis regarding competition and system paper I will submit regarding MAHED 2025 subtask1(hate and hope speech classification) in Arabic NLP colocated with EMNLP.

Time to Revisit Exact Match (Findings of EMNLP 2025)