Open nizamreplica opened 11 months ago
@nizamreplica Indeed, ClearML does not provide UI control of tasks while they are executed. You can only control a tasks configuration pre-execution ('draft') and modify descriptive elements (e.g. name, tags, description etc.) post execution.
Manually changing the tasks behaviour as it executes could lead to incoherent results, and prevent reproducibility and so is not supported. Does this make sense?
Describe the bug
If one were to change the task name under experiments during a run, the change is not reflected in the model name prefix when checkpoints are generated under models
This would make sense if the task has aborted or completed, but should work when it is running.
Note that the image attachments for the sake of this issue description are on
status=ABORTED
but the bug was found onstatus=RUNNING
To reproduce
1) Start a task that saves something as model checkpoints under models 2) While it's still running edit the task name in ClearML UI 3) Now instead of model checkpoints being saved with prefix
test_base_model_3
, it still saves them astest_base_model_2
Expected behaviour
The model checkpoints should be saved with prefix
test_base_model_3
Environment
Server type (self hosted) ClearML SDK Version: 1.13.2 ClearML Server Version: 1.12.1-397 Python Version: 3.10.13 OS (Linux, Ubuntu)