Hui Li

Achieving six times the inference throughput of current LLMs while maintaining accuracy, Nemotron 3 Ultra redefines performance benchmarks for agentic reasoning tasks.

Tencent's SIREN model boosts ad revenue by up to 3.87% in Weixin by unifying multi-modal and collaborative data into a single transformer, outperforming traditional late-fusion approaches.

Instruction tuning on a new dataset, SecGoal, allows smaller 7B/9B parameter models to outperform much larger LLMs in extracting and formalizing security goals from protocol documents.

LLM-generated rewards in RL can be misleading early in training, but RHyVE dynamically selects the best reward signal based on policy competence, leading to improved performance.

Diffusion models can now reason recursively over visual tokens, achieving state-of-the-art image generation performance by dynamically selecting specialized neural modules at each diffusion step.

Skip the retraining: AM-SGHMC lets you apply a single trained MCMC sampler to various Bayesian updating problems for similar structures.

A unified benchmark reveals the trade-offs between pixel-wise accuracy and perceptual realism in state-of-the-art image super-resolution techniques.

Don't let your SWE agent drown in context: SWE-AGILE maintains performance on multi-turn software engineering tasks by dynamically managing reasoning context with a novel sliding window and compressed reasoning digests.