`megatron_gpt_finetuning.py` does not work `max_epochs`

[AtsunoriFujita]

examples/nlp/language_modeling/tuning/megatron_gpt_finetuning.py ignores trainer.max_epochs. Always refer to trainer.max_steps only.

Tested on: nvcr.io/nvidia/nemo:24.03.framework

Test case: trainer.max_steps=200 and trainer.max_epochs=1 (187 steps) 　- 200 step job finished trainer.max_steps=200 and trainer.max_epochs=5 (935 steps) 　- 200 step job finished

Am I missing something?

[frankh077]

It is not possible to set trainer.max_epochs when you set trainer.max_steps, due to by default trainer.max_epochs is ignored.

To set trainer.max_epochs you must do it without trainer.max_steps but by default trainer.max_steps=20000 is set.

You should find a trainer.max_steps setting that suits your needs.

[github-actions[bot]]

This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.

[github-actions[bot]]

This issue was closed because it has been inactive for 7 days since being marked as stale.