🏢 Sea AI Lab

New pipeline parallelism framework achieves up to 55% higher throughput and 50% less memory usage in large language model training by systematically controlling activation memory.

Improved few-shot jailbreaking techniques efficiently circumvent aligned language models and their defenses, achieving high success rates even against advanced protection methods.

Accelerate neural network training using ‘amortized eigendecomposition’ – a novel method replacing expensive eigendecomposition with faster QR decomposition while preserving accuracy.