Code for Pre-print "Contextual Drag: How Errors in the Context Affect LLM Reasoning"

AI agents running research on single-GPU nanochat training automatically

CORAL is a robust, lightweight infrastructure for multi-agent autonomous self-evolution, built for autoresearch. Works with Claude Code, Codex, Cursor, OpenCode, Kiro, and more.

Official implementation of the paper "Escaping the Cognitive Well: Efficient Competition Math with Off-the-Shelf Models" (arXiv:2602.16793).

Codebase for the paper "How Does RL Post-training Induce Skill Composition? A Case Study Using Countdown"

[NeurIPS 2024] CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs

Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs?

[ICLR 2024] Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning

[NeurIPS 2024] SimPO: Simple Preference Optimization with a Reference-Free Reward

Collection of advice for prospective and current PhD students

[NeurIPS 2021] Multiscale Benchmarks for Multimodal Representation Learning

Reading list for research topics in multimodal machine learning