How to create config.json automatically from in-memory objects in order to archive the model?

[ynandwan]

Question
I followed the tutorial at https://allennlp.org/tutorials and at the end of it, I have a model object. I now want to archive it to be able to use later. However, archiving it (using archival.archive_model) requires a config file in the serialization directory. Is there a way to create the config file from all the objects(model, vocab etc) that I have in memory?

[joelgrus]

no, there's no way to do this. (there's no explicit specification for what a config file looks like, its structure is implicitly defined by the from_params methods, and even if you had that structure there's still be no programmatic way to extract that data from your classes).

what probably should be done is

(1) add a final section to that tutorial about how to save the weights and vocabulary (and then how to reload them), and then
(2) add a subsequent tutorial for how to train using a config file.

I guess (2) already exists as part of the README in that tutorials directory:

https://github.com/allenai/allennlp/blob/master/tutorials/tagger/README.md#using-config-files

but we need to figure out how to incorporate it into the fancy tutorial. let me open an issue for that.