Re-train your assistant

Hi,
I cannot retrain my assistant. the button is unresponsive. searched for remedy and asked friends about it. cleared cache, disabled adblockers, reset password, logged out and logged in… the problem persists. I also created a new assistant and the same issue. it just won’t work. can anyone please help?
Thanks,
Fatih

Destroy your cache on your browser. Usually a CTRL+F5 does it

:slight_smile: I already did that. Thanks anyhow. I think it is because of my custom slot types, which may be dysfunctional in the latest version of snips.

@fatih let’s investigate. When you say you cannot retrain, what button exactly are you clicking in the UI? Can you share a screenshot?

@pocketjoso
Re-train%20assistant
Thank you for your response. Also if I want to delete a value from a custom slot type, it won’t allow me. it is just seemingly deleting but it doesn’t do anything. det fungerar inte:)

@pocketjoso
If I wanna delete Clarity and Uncertainty to move them to a new Custom Slot Type named Senses, it won’t work…

the only way I get it working I created a new account and exported all the entries and imported into the new assistant/app without the values of the custom slot types… and it just created a a huge list of custom slot types with the same names… and no values in them. it works. but also I found it useful to define my own vocabulary and synonyms to train the model. I hope I am making some sense.

repetitive%20names%20custom%20slot%20types

@pocketjoso, sorry for multiple posts, forum doesn’t allow me to post multiple images in the same post. this is the other model I created in another account.

It that multiple different Slot Types with the same name? Are they for the same language? If so, this is probably the culprit. We haven’t yet had time to put in a proper error in the UI for this, but you are (no longer) allowed to have multiple resources of the same type and language with the same name. If this is the problem, re-name the slot type you want to edit into a new unique name, then try making your changes, and it should work.

Yes. The last scrshot is from the model that is working. It’s only English. When I exported from the first model, which is dysfunctional now and imported into the new model under a different user account, it created a bunch of custom slot types with the same name, “Emotions.” In the model that’s not working, there are four custom slot types with 60+ values and its synonyms. Which used to work fine.

It’s not a problem having multiple different slot types. Having multiple with the same name will just give you errors when you save, but it probably still works to train. Your training error is probably coming from something else. I’ll message you here to debug.


Multiple slot types with with multiple values won’t work. Oddly, multiple slots with the same name seems to be working. This the model has multiple slots with different names, which doesn’t work…

These are the values and synonyms for the slot type “Emotions.”

Hej Jonas,

I finally spent some time and cleaned up the extra fields in my custom dictionary strings. It works. I am wondering if there will be Swedish language support soon?

Thank you,
Fatih

Hi Fatih,
Great to hear it’s working again. In case it happens again, perhaps you had too many values somewhere? Currently our limits are:

  • Entity values, MAX 50000
  • Entity value synonyms, MAX 100 (for each value)
  • Intent training examples, MAX 5000
  • Intent training example, MAX length 1000 characters (for each training example)

Regarding support for Swedish, it’s not a language we have on our roadmap yet, so I cannot give you a specific date. It’s not likely to happen this year though.

Thank you, Jonas!

I have demoed what I made with custom intents at my current company. It is unfortunate that there’s no Swedish support. Concerning privacy issues, snips would be perfect for us.

I am curious though, if we make custom dictionaries in intents in the Swedish language and train the model with a large dataset, annotations, would that work to extract knowledge and quantitative data from unstructured data?

Glad påsk,
Fatih

I understand. As I said we will add Swedish eventually, but unfortunately it just won’t work until then.

Thank you Jonas. :smile: