teepee has quit [Remote host closed the connection]
teepee has joined #openscad
howiemnt has joined #openscad
Guest2100 has quit [Quit: quitting]
Guest2100 has joined #openscad
sculptor_ has quit [Ping timeout: 272 seconds]
sculptor has joined #openscad
sculptor_ has joined #openscad
rustgodyt has joined #openscad
sculptor has quit [Ping timeout: 245 seconds]
Guest97 has joined #openscad
Guest97 has quit [Client Quit]
rustgodyt has quit [Ping timeout: 245 seconds]
sculptor_ has quit [Ping timeout: 265 seconds]
rustgodyt has joined #openscad
<churchx>
hmm .. what might be the reasons to not allowing defining module within if(){} or difference(){}?
<churchx>
if i use module to not duplicate code for 1+ runs, not that often i need for that submodule to be used outside that scope, yet if i relocate it outside, i have to now duplicate in that outer scope scope-specific constants that are used by that submodule that is moved outside, just for Parser error about syntax to not kick in
<churchx>
it seems slightly annoying having to move submodule outside scope to calm syntax check, and so is duplicating variables (often used just to shorten code, specific to that scope only) outside those if & difference scopes
<churchx>
though maybe i just don't know possible valid rationale behind denying module definition within if/diff :)
<churchx>
moving constants more global at very least is PITA due need for them to not overlap with similar in other scopes, thus i have to add various prefixes/suffixes to generic bits like width/length/angle/offset .. resulting in overly long and cryptic constant names
<churchx>
also PITA to change them, as one needs to scroll way back and then back to scrope, due those more global constants are set often hundreds of rows away, not easy for quick tweaks when testing that scope
<Scopeuk>
If is a bit odd as it restricts scope to inside the if. I guess most people would want to do switching between two implimentations of a module which can be done inside the module. I don't know the reason for the limits. For the parameters the closest work that comes to mind would be to add parameters to the module calll so you can define the value
<rustgodyt>
Hey teepee if you're free can you check out https://filebin.net/gojf5kpn15o1l9v5 , i tried implementing Asio . Let me know if there are any issues
<teepee>
rustgodyt: that's from earlier? I had a look
<rustgodyt>
yes i sent it today morning , i could'nt see if you checked or not as i logged out
<teepee>
I wonder if maybe using boost-beast could help reducing code. If I'm reading that correctly it's header-only and directly integrates with boost.asio
<rustgodyt>
ahh yes , that is built on top of asio
<rustgodyt>
had a lot of trouble with simply asio , also a lot of extra code
<teepee>
it also seems to be well supported still and available cross platform. I did not check the packagings, but if it's really header-only, that should be not a big problem
<rustgodyt>
yeah , i'll try beast next
<teepee>
I think I've read that llama.cpp uses httplib, so that might also be an option
<teepee>
I have not yet looked at this one at all though
<teepee>
oh! "A C++11 single-file header-only cross platform HTTP/HTTPS library."
<teepee>
that's crazy :)
<rustgodyt>
right , i did come across it yesterday but i think it lacks some platform support
<teepee>
oh? https://github.com/yhirose/cpp-httplib even claims: "cpp-httplib automatically integrates with the OS certificate store on macOS and Windows. This works with all TLS backends."
<teepee>
which makes we wonder what it does on Linux but that sounds like the platform integration is not too bad
<rustgodyt>
i guess we should choose the one easiest to use
<rustgodyt>
but nowadays every library is having some security issues
<rustgodyt>
Claude mythos found this issue in this library , that issue in that library.
<rustgodyt>
so we have to check all that too
teepee_ has joined #openscad
teepee has quit [Ping timeout: 276 seconds]
teepee_ is now known as teepee
<teepee>
I don't care too much about anthropic and openai noise
<teepee>
well, except for hoping those greedy grifters collapse soon
sculptor has joined #openscad
<rustgodyt>
😂
<rustgodyt>
that aside , there are many libraries to choose from , we should choose the easiest to implement
<teepee>
yes, but I'd add well maintained to that
<teepee>
meaning the last commit is not 5 years ago
<rustgodyt>
yeah that too
<teepee>
I usually also check how many people contribute, e.g. the cpp-httplib seems to be a single person project, but having llama-cpp using it does give it bonus points :)
<rustgodyt>
i don't really know llama-cpp much , so can't bias
<rustgodyt>
although it does come on top when searching for header only cpp networking library
<rustgodyt>
i was running gpt-oss-20b:cloud with ollama in the demo, needless to say the output were pretty dumb , i said it to change the height of the candle stand - it created a box with a new height and deleted the candle stand. The output for 120b one was fine though.
<rustgodyt>
I really think i should attach some openscad docs and proper context along with the current code , if the models are running locally tokens should not be an issue , for cloud based ones the context can be modified to be a sliding window or a graph/tree based context.
<rustgodyt>
i have seen some cool plugins like graphify maybe we can do something like that
<teepee>
rustgodyt: yeah, separate topic though, also not sure if you've seen that, there's some work-in-progress to actually train a model with openscad source code modifications
<teepee>
as for running the models, if we go first with the commonly used api, the focus can be on local, but if people want to point it to somewhere in the cloud that's their option and choice
<rustgodyt>
yes that option is currently available on the demo though
<teepee>
but having some more included documentation is a great idea. at some point we added the option to ship the pdf from wikibooks but that's currently not enabled
<rustgodyt>
although i haven't tested any cloud providers
<teepee>
maybe using some markdown stuff instead might be great, even as general feature
<teepee>
I have a local ollama, but that supports the openai API too, same with llama-cpp and vllm. most likely the other one as well, notebooklm (or what's the name? I don't care too much for that due to the license)
<rustgodyt>
i can test gemini free models though , i heard they can be fine tuned so i can test fine tuning it with openscad docs
<teepee>
ah, no, that was the google thing, lm-studio is the one I meant (I think)
<teepee>
it's a first iteration, so probably not expected to be perfect by any means, but might be an interesting starting point
<rustgodyt>
Oh wait , models can be deployed in huggingface right
<teepee>
especially as the team lead is here in the channel
<teepee>
it's possible yes, I never tried that. I did try running that one on ollama
<rustgodyt>
i gotta try it then but will have to check for the completions API , does it support that ?
<teepee>
9b dense is probably not the fastest locally though, unless you really have a matching gpu
<killjoy>
`/b pdxt
<killjoy>
... stupid fingers.
<teepee>
not sure, I think so, from the description. it does expect an autocomplete style input at least
* teepee
hands killjoy some fish fingers
<rustgodyt>
the output and input styles do match in the ddescription
rustgodyt has quit [Quit: Client closed]
RoyK^ has joined #openscad
RoyK^_ has joined #openscad
RoyK has quit [Ping timeout: 265 seconds]
RoyK^ has quit [Ping timeout: 276 seconds]
<churchx>
Scopeuk: oh, i know/found out, what i need to move submodule outside. just hate all the extra work/hassle involved, especially if i don't know valid rationale behind not allowing defining module within if() or difference() (whereas eg. one can define module within other module. just not in if() & diff)
<churchx>
as result i have not 10 but hundred constants due those "moved out submodules", with less readable names due namespace exhaustion and to not overlap like cc2tipcircsidexyz, pfwthkseg1 and such and far away in beginning of file