Jonathan Giannuzzi
Jonathan Giannuzzi
It is currently possible to select multiple runs in the runs explorer of the Aim UI by clicking on each of them manually. We should add the ability for the...
When starting the server on Windows natively (so _not_ via Docker), local artifact storage is not working and prevents the server from starting. ``` > fml server time="2023-09-13T07:03:37-07:00" level=info msg="Using...
When an unknown MLFlow schema version is encountered, we should bail out and let the user know that they can use `--force` to try the import, _without any guarantees of...
We should first determine a safe range of schema versions across which the basic structure of the objects we deal with has not changed. I would suggest from `c48cb773bb87` to...
For this MVP, we will only support the latest MLFlow schema at the time of implementation, without any `--force` flag. We _do_ need to make sure we don't run any...
Add the ability to import an existing MLFlow database — _without any migrations or changes to it_ — into an existing or new FML database. This will be a continuation...
v5 was introduced in Vault 1.6. We are currently using v4, which is still supported but will be removed in a future release. See the [upgrade guide](https://github.com/hashicorp/vault/blob/38b2375ca4cb9d10f8441a5fa581f762703fde85/sdk/database/dbplugin/README.md) for more details.
We should provide an `import` command that allows to import an existing MLFlow database (sqlite or postgres) into an FML database. When opening an existing an MLFlow database, we currently...
Once #50 is implemented, the list of available param/metric names should be limited to the ones used by the current experiment.
Once we have added [support to visualize artifacts in the Aim UI](https://github.com/G-Research/fasttrackml/issues/228), we should add a way to see a diff between 2 artifacts from different runs. This should be...