[1092] Using MLFlow in the main repo #1127
Conversation
| readme = "../../README.md" | ||
| requires-python = ">=3.12,<3.13" | ||
| dependencies = [ | ||
| "mlflow", |
There was a problem hiding this comment.
maybe mlflow-skinny would work
| channels_set = collect_channels(scores_dict, metric, region, runs) | ||
|
|
||
| for run_id, metrics_dict in reordered_dict.items(): | ||
| parent_run = get_or_create_mlflow_parent_run(mlflow_client, run_id) |
There was a problem hiding this comment.
the problem is the run_id here, which is the inference run.the easiest is probably to query mlflow to find the value of the tag tags.from_run_id (which will be the model).
This or reading the config for this run. I think it will be more complicated.
There was a problem hiding this comment.
I pushed an update, getting from_run_id from the run config and using it to generate parent_run.
Upon uploading the scores, ordering on MLFlow looks as below:
We have to check if everything else is in order in terms of relations between run_ids.
There was a problem hiding this comment.
Not sure if we should keep the inference run_id in the name or rather have it in the metadata. On the other hand might be good to keep it in the name in case of several inference instances...
Description
Issue Number
Refs #1092
Is this PR a draft? Mark it as draft.
Checklist before asking for review
./scripts/actions.sh lint./scripts/actions.sh unit-test./scripts/actions.sh integration-testlaunch-slurm.py --time 60