earonesty

@vitorpamplona i like your solution, except then the room's participants are public, which might not be a good thing. i think simply rolling a new room every time people enter/leave...

llama.cpp seems to support this now fyi!

i ran into this problem. NamedTemporaryFile should never be used on windows or linux. it is fully bugged and can even result in an an infinite loop. uuid mechanism is...

i think the issue is you now need to specify the chat_format correctly ... it won't guess anymore.

ok, got this to work too. i downgraded watermelon to < 0.26, and it works now new error looks like application logic issue, we get 100's of these ``` 05-26...

ok, i wrapped it in a async-mutex, that fixed the console spew and finally gives a consistent view of the db (but of course now it's super slow). going to...

the sync api seems to rely on watermelon dictating the ids. unfortunately, i can't use those id's. our server and schema is highly constrained, and we must use the id's...

you might want to try this: https://github.com/autocyz/pytorch2ncnn. cuz theoretically, you should be able to autoconvert. pytorch models are just code, and so is c++/ncnn. so a codegen pipe should work

PositionInAlignment _should_ point to AlignedBases (and i think it does) so you can look at deletions piled-up as well. I created a "positionInQueryOffset" and increment it as it walks through...

digest comes with ruby. it exists. i think maybe an older ruby might work.