🏢 National University of Singapore

LongPO: Self-evolve LLMs to excel in long contexts via short-to-long preference optimization, boosting performance without sacrificing short-context skills.

NExT-Mol: Combines 1D language models with 3D diffusion for molecule generation, achieving state-of-the-art performance and validity.

CoT-Valve dynamically adjusts reasoning chain lengths based on task difficulty, significantly reducing inference costs in large language models without substantial accuracy loss.

Enhance-A-Video boosts video generation quality without retraining, by enhancing cross-frame correlations in diffusion transformers, resulting in improved coherence and visual fidelity.

GuardReasoner enhances LLM safety with reasoning-based guardrails, improving performance, explainability, and generalization on various benchmarks.

CLEAR: Conv-Like Linearization boosts pre-trained Diffusion Transformers, achieving 6.3x faster 8K image generation with minimal quality loss.

TinyFusion, a novel learnable depth pruning method, crafts efficient shallow diffusion transformers with superior post-fine-tuning performance, achieving a 2x speedup with less than 7% of the original…

Collaborative Decoding (CoDe) dramatically boosts visual auto-regressive model efficiency.

OminiControl: A minimal, universal framework efficiently integrates image conditions into diffusion transformers, enabling diverse and precise control over image generation.

AnchorAttention enhances long-context LLMs by mitigating BFloat16’s disruptive effects on RoPE, improving performance and speeding up training.

Boost large language model training speed by 51% with Vocabulary Parallelism, a novel technique that balances computation and memory usage across pipeline stages.

GenXD: A unified model generating high-quality 3D & 4D scenes from any number of images, advancing the field of dynamic scene generation.