How to change the upper level config?

[Hannibal046]

Discussed in https://github.com/facebookresearch/hydra/discussions/2651

^{Originally posted by **Hannibal046** April 30, 2023} Hi, Thanks so much for this great projects! I am relatively new to Hydra, and I have a problem about **changing the upper level config**. Suppose that my config folder looks like this: and `conf/config.yaml` looks like this: What i want is for `wmt` dataset, the `max_epochs` for 100 while for `iwslt` dataset, the `max_epochs` is 10. How could i implement this? I know [Specializing configuration](https://hydra.cc/docs/patterns/specializing_config/), but this seems to handle a special case where there is only one should be specialized. But in my case, i may have much more datasets and i want to give each of them different trainer values, how could i achieve this?

[Hannibal046]

I used to employ lightning cli to manage my configs and it simply use config overriding to handle this.

For example, i would have a general_config.yaml like this:

And i have a dataset specific config dataset.yaml like this:

When executing my command, the trainer arguments in dataset.yaml would override those in general_config.yaml. Do we have similar solutions in Hydra?

python main.py --config general_config.yaml --config dataset.yaml

[omry]

Typically, the configs in a config group are configuring the name with the same config group. If you want to modify the top level config, you can use a global package in the configs. This is similar what is described here.