Skip to content

πŸš€ Introducing scale and spatial Markovian assumptions into VAR β€” achieving faster generation with reduced memory consumption.Β #164

@NuanBaobao

Description

@NuanBaobao

Thanks to the great work on VAR! πŸ™Œ
Building upon it, we introduce MVAR, which incorporates scale and spatial Markovian assumptions into visual autoregressive modeling.
MVAR achieves a 1.7Γ— speedup and 3Γ— reduction in GPU memory usage, enabling efficient training on eight RTX 4090 GPUs.

πŸ“„ Paper: https://arxiv.org/abs/2505.1274
πŸ”— GitHub: https://github.com/LabShuHangGU/MVAR

MVAR Overview

Metadata

Metadata

Assignees

No one assigned

    Labels

    ResearchThird-party repositories or research which use VAR

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions