Skip to content
This repository has been archived by the owner on Nov 22, 2022. It is now read-only.

Improve get_logits() #683

Closed
wants to merge 1 commit into from
Closed

Improve get_logits() #683

wants to merge 1 commit into from

Commits on Dec 9, 2019

  1. Improve LogitsWorkflow to handle dumping of raw inputs and multiple o…

    …utput tensors (#683)
    
    Summary:
    Pull Request resolved: #683
    
    1. Dump model inputs if configured to do so. This makes preparing the input file for KD easier: an extra step to combine logits file and input file is not required.
    2. Check for existence of file in squad.unflatten() and print a message but don't fail. PyText loading has a limitation that for task creation tries to load the train/test/eval files in model config. However, the files may not exist at the time of loading the model. To overcome the limitation adding a check to ignore files that don't exist. The drawback of this is that during model training if the file doesn't exist then the training will not fail but go on. However, if this occurs then the model will not train at all and the metrics should raise a flag. Plus there's a warning message that flags the fact file paths supplied did not exist.
    
    Reviewed By: borguz
    
    Differential Revision: D15757521
    
    fbshipit-source-id: f1271b10d77e5894581af2781a550a254f1bd4e2
    Kushal Lakhotia authored and facebook-github-bot committed Dec 9, 2019
    Configuration menu
    Copy the full SHA
    a4861fd View commit details
    Browse the repository at this point in the history