[07:57:03] <elukey>	 kevinbazira: o/ wow 1.56G -> 800M is great! 
[08:06:43] <kevinbazira>	 yep, it is. Thanks to your proposal elukey! :)
[08:07:50] <elukey>	 \o/
[12:24:30] <klausman>	 Nice work
[13:12:35] <elukey>	 added some changes to the kubeflow-kfserving-inference chart: https://gerrit.wikimedia.org/r/c/operations/deployment-charts/+/719515/1/charts/kubeflow-kfserving-inference/README
[13:13:20] <elukey>	 I tried to split the configs of InferenceServices between "base" and "custom"
[13:13:30] <elukey>	 so that it should be easier when deploying new models
[13:13:51] <elukey>	 the helmfile config for the first model is
[13:13:52] <elukey>	 https://gerrit.wikimedia.org/r/c/operations/deployment-charts/+/719128/4/helmfile.d/ml-services/revscoring-editquality/values.yaml
[13:13:58] <elukey>	 (based on the above chart change)
[13:14:16] <elukey>	 so the amount of copy/paste for new models should be minimal
[13:29:36] <wikibugs>	 10Lift-Wing, 10Machine-Learning-Team (Active Tasks): Implement Pod Security Policies for Istio/Knative/Kfserving - https://phabricator.wikimedia.org/T288829 (10elukey) Thanks to @JMeybohm's refactoring in https://gerrit.wikimedia.org/r/c/operations/deployment-charts/+/719295 we'll be able to solve the knative/...
[15:08:22] <chrisalbon>	 morning all
[15:55:03] <elukey>	 chrisalbon: o/
[15:55:59] <elukey>	 accraze, kevinbazira - qq about how to store models on swift: do we have a short/medium plan about how to upload the models until we have something more automated? (gitlab etc..)
[15:56:13] <elukey>	 I didn't find a task for it, in case I can open one
[15:56:45] <elukey>	 we could even simply use stat100x hosts for the moment (in case I'll need to do some tweaks to the firewall to allow traffic to swift to pass
[15:56:48] <elukey>	 )
[16:21:55] <accraze>	 o/
[16:22:28] <accraze>	 elukey: i was thinking about storing models on swift yesterday...
[16:23:11] <accraze>	 the past two models i've just done on ml-serve using a script i wrote
[16:23:23] <accraze>	 similar to this: https://phabricator.wikimedia.org/P15960
[16:23:53] <accraze>	 following the format outlined in:  https://phabricator.wikimedia.org/T280467
[16:24:48] <accraze>	 stat100x would work too
[16:29:58] <accraze>	 lets discuss during tech meeting today
[16:35:19] <elukey>	 accraze: ack perfect! 
[18:00:53] * elukey afk!
[21:24:45] <wikibugs>	 10Lift-Wing, 10Machine-Learning-Team (Active Tasks): Naming convention for the model storage structure - https://phabricator.wikimedia.org/T280467 (10ACraze) Echoing @klausman and @calbon from our ML Team Technical Meeting today:  Let's go with seconds-granularity for now when creating a timestamp version. The...