Cornstarch: Efficient Distributed MLLM Training with Frozen-Aware Parallelism and Balanced Context Parallelism
Model parallelization with two types of modality parallelism and pipeline parallelismAbstract
Review of paper writing approach, summarizing research themes and innovations, sharing paper submission experiences.
Date
Sep 30, 2026 8:30 AM — 9:15 AM
Event
Cornstarch: Efficient Distributed MLLM Training with Frozen-Aware Parallelism and Balanced Context Parallelism
Location
Room 101, Xinda Building B
Authors
Hao Yao
(he/him)
M.S. Student
Master’s student at CUIT. Research focuses on optimization of model and data heterogeneity in large-scale training of Multimodal Large Language Models (MLLM).