QUESTION about the performance when the amount of sentences increase

[tobiloeb]

Current I am playing around with home assistant voice. I love to talk to my home. Control my devices and ask for sensors and statuses in my smart home. And with the defined sentences here it's very fast in my HA setup and I don't have to spend tokens to Gen AI for it. ;) So thank you for your work here!!

I can imagine that I will create some more PR here in the near future. And that's why this question comes up to my mind: What's the impact of the performance when I (or others) add more and more sentences?

Are there any performance tests or experiences how it effects the performance when the amount of sentences increases? So should we/I balance (weigh up) which sentences will be added?

Regards, Tobi

[hecktech27]

https://github.com/OHF-Voice/intents/pull/3268#issuecomment-3095724580 This comment from @mib1185 on one of my PRs might be as interesting for you as it was for me. have a look at those "old" numbers and then look at todays sentence count 😉

[tobiloeb]

@hecktech27 Oh wow didn't found this, thanks for sharing. 👍

Currently we have about "only" 15.000.000 sentences for DE. 😄 So I can add some more without worries. Do you know whether there are a threshold we shoudn't reach?

[hecktech27]

To be honest i don´t know. There has been quite some effort during the last months for the DE-sentences to bring down sentence count to where it is right now(without loosing tested sentences). For me it has become good practce to always check sentence count before and after my changes and if the increase felt unreasonably high I double checked if e.g. one of the expansion rules was adding unnecessary permutations to the sentences. But since we are (given the complexity of the german language) extremely low on sentence count I personally think there is some headroom to add new functionality. Maybe @andreasbrett can share some more insight on this since I´m running HA on a quite powerfull machine and honestly did not realize that much of a difference.

[andreasbrett]

No there's no such thing as a threshold. We should not return to old times where sentence count wasn't considered though. At the time the performance issues were addressed, Mike Hansen implemented several things to improve the situation including a caching system. So it is unclear which measure provided the best performance boost. Also I remember that French performed much better than German, though it had a higher sentence count.

Nevertheless, I would like to keep sentence count as an important metric. We shouldn't restrict ourselves too much though and keep a good balance of performance and innovation. So when there's a good reason to introduce a lot of sentences, then it's fine. But it should be accounted for.