Jump to content

18492492. TRAINING OBJECTIVES FOR DISTILLING GUIDED DIFFUSION MODELS (QUALCOMM Incorporated)

From WikiPatents

TRAINING OBJECTIVES FOR DISTILLING GUIDED DIFFUSION MODELS

Organization Name

QUALCOMM Incorporated

Inventor(s)

Risheek Garrepalli of San Diego CA US

Shubhankar Mangesh Borse of San Diego CA US

Jisoo Jeong of San Diego CA US

Qiqi Hou of San Diego CA US

Shreya Kadambi of San Diego CA US

Munawar Hayat of San Diego CA US

Fatih Murat Porikli of San Diego CA US

TRAINING OBJECTIVES FOR DISTILLING GUIDED DIFFUSION MODELS

This abstract first appeared for US patent application 18492492 titled 'TRAINING OBJECTIVES FOR DISTILLING GUIDED DIFFUSION MODELS

Original Abstract Submitted

A method for training a diffusion model includes compressing the diffusion model by removing at least one of: one or more model parameters or one or more giga multiply-accumulate operations (GMACs). The method also includes performing guidance conditioning to train the compressed diffusion model, the guidance conditioning combining a conditional output and an unconditional output from respective teacher models. The method further includes performing, after the guidance conditioning, step distillation on the compressed diffusion model.

Cookies help us deliver our services. By using our services, you agree to our use of cookies.