So, the 1st image (jupyter) is the code I used to create a task and upload the artifact. Then the task was created and I can access it normally with web GUI (like in the 2nd image). But when I click "artifacts" tab, the 404 error page appears (3rd image).

Thanks AgitatedDove14 , I tried setting TEMP environment variable to another directory and it worked! Not sure whether TMP could work too.

I have ClearML set up locally. The way to run the task is straightforward: I create the task with Task.init() at the very top of the file, do things (inference, save outputs, etc.), upload outputs with task.upload_artifact(), and then end the script with task.mark_completed().

SuccessfulKoala55 Sure will do. :)

I see. That's quite handy. Thanks!

SuccessfulKoala55 I see. Hope it will be added as a new feature in the future version. For me, it's quite important for the organization purpose, especially if the task outputs many artifacts.

SuccessfulKoala55 I am trying to find a way to work around it for the time being. I have 2 requirements: 1) I want to log a custom metric that is computed only at the end of every epoch (unlike other tf metrics which are updated per mini-batch). If I follow the tf doc here, will ClearML log it for me and show on "scalars" tab? https://www.tensorflow.org/tensorboard/scalars_and_keras#logging_custom_scalars 2) It's the same thing as 1) but it's a image instead. https://www.tensorflow.org/te...

CostlyOstrich36 Yes, the same thing occurs. version: 1.1.2 for clearml-agent

It printed "ClearML couldn't detect iterations..." something like this, but the process never ends, just freezes here. I mean I can't run any further commands with this terminal cuz it's running this task, except ctrl+c.

SuccessfulKoala55 I don't think so cuz the files are just small dataframes and the thing is I tried saving those output files on my local machine then created a new task, uploaded them with a new code on .ipynb, and it took less than a minute and everything works fine. (the frozen script is .py)

I didn't call task.execute_remotely() so it didn't push the task to the workers