Siva Reddy

The Markovian Thinker: Architecture-Agnostic Linear Scaling of Reasoning

Value Drifts: Tracing Value Alignment During LLM Post-Training

MMTEB: Massive Multilingual Text Embedding Benchmark

The Promise of RL for Autoregressive Image Editing

Not All Data Are Unlearned Equally

VinePPO: Refining Credit Assignment in RL Training of LLMs

SafeArena: Evaluating the Safety of Autonomous Web Agents

AgentRewardBench: Evaluating Automatic Evaluations of Web Agent Trajectories

BigDocs: An Open Dataset for Training Multimodal Models on Document and Code Tasks

VinePPO: Unlocking RL Potential For LLM Reasoning Through Refined Credit Assignment

How to Get Your LLM to Generate Challenging Problems for Evaluation