How do I build the nocharlm pos, lemma, and depparse models?

[rhdunn]

I'm using the stanza.utils.training.run_[model_type] entry points to build the models. I'm currently just passing --save_dir, --save_name, and --train/--score_test -- except for pos and depparse where I'm also passing --wordvec_pretrain_file.

This is building the charlm variants of the models. With the update to Stanza 1.6, the lemma model is now building with charlm data.

Is there a way to build the nocharlm variants?

The --charlm option appears to be present to turn charlm on, but the behaviour seems to be that this is on by default given the resulting file sizes and the need to specify the forward/backward charlm models in the resources.json file.

The --no_char option looks as though it might disable charlm, but is only present for the pos (tagger) and depparse (parser) models. The argument documentation is also confusing as this option refers to "character model" where the charlm options say "character-level language model".

[AngledLuffa]

`--no_charlm` `char` refers to the built in character model. a little confusing, i know

…

On Tue, Nov 14, 2023, 12:08 PM Reece H. Dunn ***@***.***> wrote: I'm using the stanza.utils.training.run_[model_type] entry points to build the models. I'm currently just passing --save_dir, --save_name, and --train/--score_test -- except for pos and depparse where I'm also passing --wordvec_pretrain_file. This is building the charlm variants of the models. With the update to Stanza 1.6, the lemma model is now building with charlm data. Is there a way to build the nocharlm variants? The --charlm option appears to be present to turn charlm on, but the behaviour seems to be that this is on by default given the resulting file sizes and the need to specify the forward/backward charlm models in the resources.json file. The --no_char option looks as though it might disable charlm, but is only present for the pos (tagger) and depparse (parser) models. The argument documentation is also confusing as this option refers to "character model" where the charlm options say "character-level language model". — Reply to this email directly, view it on GitHub <#1308>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA2AYWO3MBIW2L4BJAO7LK3YEPFVDAVCNFSM6AAAAAA7LMGH3CVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZVHA2TCOJWGY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[rhdunn]

Thanks! That worked.