search

WildMeOrg / scout

MIT License
6 stars 5 forks source link

Reduce Default for WIC_BATCH_SIZE #49

Open TanyaStere42 opened 1 year ago

TanyaStere42 commented 1 year ago

Current Behavior Scout auto-annotation ML (v1, v2) appears to run on 4GB NVIDIA 1050ti, tells you it’s completed the annotation (100%, ready for truthing) but doesn’t actually annotate anything. Testing on command line in same container with scoutbot batch throws errors about ‘unable to allocate memory’. Playing with export WIC_BATCH_SIZE=160 (down from the default 256?) leads to success.

Desired Behavior Set export WIC_BATCH_SIZE=160 down from the default 256

simbamangu commented 1 year ago

I think this should be labelled as a bug, given that it reports finished annotation but doesn't actually do it.