One-Step Gradient Delay is Not a Barrier for Large-Scale Asynchronous Pipeline Parallel LLM Pretraining
Paper • 2606.30634 • Published • 18
Andrey Veprikov
Veprikov
AI & ML interests
Optimization
Recent Activity
upvoted a paper about 3 hours ago
One-Step Gradient Delay is Not a Barrier for Large-Scale Asynchronous Pipeline Parallel LLM Pretraining upvoted a paper 5 months ago
Quartet II: Accurate LLM Pre-Training in NVFP4 by Improved Unbiased Gradient Estimation upvoted a paper 7 months ago
Kandinsky 5.0: A Family of Foundation Models for Image and Video GenerationOrganizations
None yet