Possible to override tokenizer at spacy.load function via config?

[kinghuang]

The spacy.load function takes a config argument that allows the model config to be tweaked at load time. Is there a way to override the nlp.tokenizer config? I'm trying to do the following, but getting an error that the section value cannot be overwritten.

nlp = spacy.load("en_tsds_doc_maker", config={"nlp.tokenizer": {"@tokenizers":"oi.Tokenizer.v1"}})

confection.ConfigValidationError: 

Error parsing config overrides
nlp -> tokenizer -> @tokenizers	not a section value that can be overwritten

What I want to do is assign {"@tokenizers":"oi.Tokenizer.v1"} to nlp.tokenizer. But, it appears to be trying to assign oi.Tokenizer.v1 to nlp.tokenizer.@tokenizers. Is there some way to make this work?

[bdura]

Hello @kinghuang, that's a good question - we'll need to look into what exactly is happening there.

However, in general overriding the tokenizer configuration is not a great idea because it's unlikely you'll be able to load the model in from_disk.

[kinghuang]

Thanks. In this case, the alternate tokenizer is packaged with the model. I can build two separate model packages, but I'd rather build one and just swap out the tokenizer on the fly via config.

[kinghuang]

Fixed in #12623.