CLI parsing error: fit -c run.yaml works but -c run.yaml fit fails

[oooooio]

Bug description

When running Lightning CLI with a config file, the position of the subcommand (fit, validate, etc.) affects parsing.

• ✅ Works:

uv run src/trainer.py fit -c run.yaml

• ❌ Fails:

uv run src/trainer.py -c run.yaml fit

• Error

usage: trainer.py [-h] [-c CONFIG] [--print_config[=flags]] {fit,validate,test,predict} ...
error: Validation failed: Key 'model.backbone.init_args.seq_length' is not expected

• Expected behavior
  Both fit -c run.yaml and -c run.yaml fit should behave identically, or the CLI should raise a clearer error message.

What version are you seeing the problem on?

v2.5

Reproduced in studio

No response

How to reproduce the bug

Error messages and logs

# Error messages and logs here please

Environment

Current environment

#- PyTorch Lightning Version (e.g., 2.5.0):
#- PyTorch Version (e.g., 2.5):
#- Python version (e.g., 3.12):
#- OS (e.g., Linux):
#- CUDA/cuDNN version:
#- GPU models and configuration:
#- How you installed Lightning(`conda`, `pip`, source):

More info

No response

cc @mauvilsa

[bhimrazy]

Hi @oooooio, thanks for reporting this.
Could you please share the full error trace and, if possible, a minimal reproducible example?

[mauvilsa]

This is a duplicate of #19714. It is expected that giving the config before or after the subcommand has different behaviors. Similar to the output of cli.py --help and cli.py fit --help being different. What should improve is the error message. Not strictly a bug.

[oooooio]

This is a duplicate of #19714. It is expected that giving the config before or after the subcommand has different behaviors. Similar to the output of cli.py --help and cli.py fit --help being different. What should improve is the error message. Not strictly a bug.

it's true