4 papers
MuonBP: Faster Muon via Block-Periodic Orthogonalization
Ahmed Khaled, Kaan Ozkara, Tao Yu +2
Gradient orthogonalization is a simple strategy that shows great utility in speeding up gradient descent. The Muon optimizer (Jordan, Jin, et al., 2024) combines gradient orthogona…
Stochastic Rounding for LLM Training: Theory and Practice
Kaan Ozkara, Tao Yu, Youngsuk Park
As the parameters of Large Language Models (LLMs) have scaled to hundreds of billions, the demand for efficient training methods -- balancing faster computation and reduced memory…
Training LLMs with MXFP4
Albert Tseng, Tao Yu, Youngsuk Park
Low precision (LP) datatypes such as MXFP4 can accelerate matrix multiplications (GEMMs) and reduce training costs. However, directly using MXFP4 instead of BF16 during training si…
Collage: Light-Weight Low-Precision Strategy for LLM Training
Tao Yu, Gaurav Gupta, Karthick Gopalswamy +7
Large models training is plagued by the intense compute cost and limited hardware memory. A practical solution is low-precision representation but is troubled by loss in numerical…