260 karma · joined June 7, 2021
or the battery, it does not like high temperature either.
you only regret it when it happens. and it will happen.
There are a lot of jobs that should run in a home server running 24/7 instead of abusing your poor laptop. Remote dedicated servers work, but the latency is killing your productivity, and it is pricey if you want a server with a lot of disk space.
Open weight models are still useful to extract knowledge from. With much lower cost compared to hitting the API
The point is pretty simple, if the author is too lazy to do some basic research about this common problem, he is probably too lazy to do other important stuff, too.
Time is short, there is always new languages out there to test.
make me instantly lost interest in the language.
You have already killed ten millions people and you still do not have enough?
How much bloodthirsty you are?
Sure, for the demo they can choose postmark as a mockup server. But he should keep in mind that postmark is not a hard requirement, eventually they need to switch to something more controllable/cost effective.
Failing to knowledge that is already a considered failure.
Look at all the third party services he uses! Nobody sane would pick that solution. That would be a completely maintainment nightmare.
He did not write an email client, he wrote a wrapper to some email client, which handles all the heavy lifting.
At the very least, write something that interact with a real (or fake) SMTP server. Other things are just cherry on top.
sorry, but to me it is just pure garbage.
Why not just compress the whole database using `gzip` or `lz4` before rsyncing it instead? `zstd` works too but seems like it had a bug regarding compressing file with modified content.
better yet, split your sqlite file to smaller piece. it is not like it needs to contain all the app data in a single sqlite file.
will never happen.
it annoys me to no end that ever americans out there always assume that the war was between north and south vietnam, while in reality it was not. VCs were mostly southern people, the conflicts were mostly between south vietnam and the National Liberation Front of South Vietnam which was found and led by southern vietnamese.
which mean the conflict would never stop, even without the help from north vietnam or soviet.
before using crystal I mostly used Ruby for scripting, but there were a lot of problems - the program often crashes after running a while due to typo/nil handling, causing a lot of time wasting - it was too slow for my case, which involving with dealing with large amount of data - doing things in parallel (fetching websites, parsing multiple files) is a lot of work, you need to use external gems for the task, and it is not always the way you want it to do
there is a lot of works spent on crystal to make it enjoyable, you almost never need external libraries to do stuffs. the standard library is big and helpful.
I'm currently using github copilot with claude 3.7, mostly for auto complete, work pretty well although sometimes it does return ruby variant of the code (it call the ruby methods that not existing in crystal)
but for my use case, copilot works pretty well.
> long running programs depends, I write several services that running 24/7. Some runs fine for months, some just refuses to response after a few day. Too busy to track for the cause so I just automatically reset the affected services at 4AM.
At the very least the conversion between the versions will be very messy and error prones, especially using the migration tools.
Not to mention the designs of runes was still unstable back then. Heck even now it is still very awkward to use runes with sveltekit. You need $state and $effect everywhere for the page data.
Without svelte 4 compatibility, you can just replace `$state` with `$`, `$derived` with `$$` or even some simpler mechanisms. Heck, I have vague intuition that we can reuse the `$: ` statements someway too, and still keep the new deep reactivity feature.
Might be svelte 6 will give us the proper answer. Or some newer fancy framework will take the hint and do it sooner.
Do you know why there are new frameworks, new versions every time? Because web is a moving target. New js/css/html5 features are released every day and the libraries have to update to adopt them.
Back in the day you do not have css variables, so you have either use sass and do all the math inside the preprocessing process, or crazier, using those shitty css-in-js libraries to calculate css properties on the fly. Nowadays you can pass around the variables and it can cover almost all the case. In able to pass the css property to the component, you need to improve your framework.
Another example is html5 history API, back in the old days people implemented link components in SPA with buttons (hilarious but it happened), only after History API is introduced there raised the need to manipulate the history properly, thus another rewrite issued.
before that, ES5 introduced `class`, proving yet another way to write js code, which in some situation make cleaner code than the plain old js functions.
everything in the web is moving, new nice things are introduced every month, if you want to create a modern and performant web application, you need to keep up with the new fancy things, frameworks are no exception for that.
No, I want GPU. BERT models are still useful.
The point is your service is too expensive that only one or two months of renting is enough to build a PC from scratch and place it somewhere in your workplace to run 24/7. For applications that need GPU power, usually downtime or latency does not really matter. And you always add an extra server to ensure.
Like everything. You do not even mention your machine spec, so I'm assume you just pick the ones that fit, which probably the quant versions.
Quant versions of the "small" models do not perform that well. Not the way you expected them to be.