fix(wandb): log different val datasets separately in wandb

[grgkovac]

What does this PR do ?

Fixes a bug with logging multiple val datasets to wandb. They all write in Charts/val_loss, and the last val dataset overwrites the previous ones. This fix logs different val datasets in different sections. See screenshots below.

Changelog

• lines L1541-1545 in nemo_automodel/recipes/llm/train_ft.py
• L1541 if val_name == "default", only one dataset is used without a specific name (validation dataset: in the yaml) -> same as before
• L1544-1545: get val_name and use it as section name.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests? not needed
• Did you add or update any necessary documentation? not needed

Additional Information

Two dummy val datasets are used for testing: dummy_number dummy_letters.

After the fix - two section one dummy_letters one dummy_numbers

Before the fix - one val_loss in Charts (dummy_letters under the name val_loss, dummy_numbers not logged)

After the fix one val dataset with custom name (validation_dataset_dummy_letters: in yaml) - separate section

After the fix one val dataset with default name (validation_dataset: in yaml) - val_loss in Charts as before

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[akoumpa]

/ok to test 1db9ea9

[grgkovac]

@akoumpa the L2_HF_PEFT failure looks unrelated to this PR. Do you mind rerunning the tests?

[akoumpa]

/ok to test bfbe695

[akoumpa]

Thanks @grgkovac ; restarted CI now.