Skip to content

Multi-label Bert Classification with focal loss weighting, auto cross-label data synthesis, adding exclude loss part among specific labels, upsampling, robust mean over all positive or negative loss, generating very fast inference-time model, etc.

Notifications You must be signed in to change notification settings

qf6101/multi-label-bert-classification

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Multi-label Bert Classification

This implementation adds useful features on bert classification:

  1. Multi-label
  2. Focal loss weighting
  3. Auto cross-label data synthesis
  4. Adding exclude loss part among specific labels
  5. Upsampling
  6. Robust mean over all positive or negative loss
  7. Generating very fast inference-time model

N.B. I deleted the efficient model service part (about 500 qps per 1080ti gpu).

About

Multi-label Bert Classification with focal loss weighting, auto cross-label data synthesis, adding exclude loss part among specific labels, upsampling, robust mean over all positive or negative loss, generating very fast inference-time model, etc.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published