Make base-config flag exclusive to fresh training, allow empty

[grassesi]

Description

Issue Number

closes #1539

Is this PR a draft? Mark it as draft.

Checklist before asking for review

• I have performed a self-review of my code
• My changes comply with basic sanity checks:
  • I have fixed formatting issues with ./scripts/actions.sh lint
  • I have run unit tests with ./scripts/actions.sh unit-test
  • I have documented my code and I have updated the docstrings.
  • I have added unit tests, if relevant
• I have tried my changes with data and code:
  • I have run the integration tests with ./scripts/actions.sh integration-test
  • (bigger changes) I have run a full training and I have written in the comment the run_id(s): launch-slurm.py --time 60
  • (bigger changes and experiments) I have shared a hegdedoc in the github issue with all the configurations and runs for this experiments
• I have informed and aligned with people impacted by my change:
  • for config changes: the MatterMost channels and/or a design doc
  • for changes of dependencies: the MatterMost software development channel

[clessig]

@Jubeku : you had a case where you wanted to use base-config with train_continue. Why was a new base-config necessary and one could not solve it with overrides?

[Jubeku]

@Jubeku : you had a case where you wanted to use base-config with train_continue. Why was a new base-config necessary and one could not solve it with overrides?

Yes, see #1663

[clessig]

But why do you need a new base config and cannot just overwrite the relevant parameters. Maybe you can be more specific in the other issue you created

…

________________________________ From: Julian Kuehnert ***@***.***> Sent: Wednesday, January 21, 2026 9:58:20 AM To: ecmwf/WeatherGenerator ***@***.***> Cc: Christian Lessig ***@***.***>; Comment ***@***.***> Subject: Re: [ecmwf/WeatherGenerator] Make base-config flag exclusive to fresh training, allow empty (PR #1661) [https://avatars.githubusercontent.com/u/23659747?s=20&v=4]Jubeku left a comment (ecmwf/WeatherGenerator#1661)<#1661 (comment)> @Jubeku<https://github.com/Jubeku> : you had a case where you wanted to use base-config with train_continue. Why was a new base-config necessary and one could not solve it with overrides? Yes, see #1663<#1663> — Reply to this email directly, view it on GitHub<#1661 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AHCHOHVLXX6INZGAFVKIG4L4H45SZAVCNFSM6AAAAACSJF62RWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTONZWHEZTANBRGE>. You are receiving this because you commented.Message ID: ***@***.***>

[Jubeku]

#1663 (comment)

[sophie-xhonneux]

It would be good for fine-tuning experiments to be able to override the base-config, so that model_inputs, etc can be removed without knowledge of what they were during pre-training

[grassesi]

All the usecases mentioned here, can be achieved using the --config flag. Indeed this was the intention of this flag. Care has been taken that every parameter is overwritable. Using the --config flag it is much clearer what actually has been changed specifically for finetuning, it also allows for more modularity: A specific fintuning override could be used for different base configs. Abusing the --base-config flag for this usecase will only lead to more copy-pasting around base configs and pontential errors in doing so. If there anything you want to do that --config is preventing you from doing, please let me know and I am happy to find a solution.

[simone99n]

Was this PR replaced by PR#1738?

[grassesi]

Was this PR replaced by PR#1738?

No this is complementary to #1738 but less urgent