machinekob

Sorry here in EU we are making a lot less past 2 years we have big increase in salary but still avg for EU is probably about ~60-80k USD if you have 4/5+ years of exp and if you are very good you can get 100-120k USD+ (not counting Meta/Google and few other but only in few countries) and ofc it is pre-tax as taxation is a lot higher than US so in some countries you'll get about 40-60% total salary (counting VAT on basic products and other taxation).

Not even counting Eastern Europe and Asia where you get fraction of that.

Nvidia smells a lot of profit and maximise it by forcing prosumers to buy trashy value "quadro" or even worse buying data center GPU's for working on state of the art problems (6-20k usd minimum investment to do normal sized Deep Learning in next 2/3 years incoming and probably 3/4x that for training).

Big fuck you for nvidia with this release prosumer market is dying right now and there is no alternative cause AMD software is so bad you are forced to use CUDA in for example DL/DS.

2be honest numpy is ez to install on all major platforms. In deep learning I'm almost never saw usage of pandas but deep learning models have problems with PyTorch as some projects just lock old PyTorch version that just dosent work on new/old python version.

There isn't any optimised diffuser on m1 yet most of them are just running basic MPS graph or even mix of CPU and MPS ops and ofc it is extremely slow I dont have time to test this implementation but author just use some other implementation with simple UI. So I would be surprise if it is faster than 10s per img with 20steps and probably more close to 25-40s and around 40s-1min per classic 512x512 50-60 step setting as are other models.

CUDA vs MPS happen as apple software is always behind hardware by a lot.

The problem is mostly about model training and architecture. I was doing TTS like 2.5/3 years ago and most models were train on fixed (+/- 5-10s) clips with like avg of 80 words or so, there were few attempts for fixing that and if I remember correctly few RNN-based models were good at ignoring input length and generate "good" audio but new flow-based and diffusion based models are out of my domain as I'm in CV for past few years and only read some new cool paper once in a while :)

You can also search for postags (and token ids for them) that are especially placed for "pause" audio as they often fix problem with weird transition when you split the sentences.

This repo -> https://github.com/TensorSpeech/TensorFlowTTS was very good few years back.

But you need new motherboard and worst of all its only DDR5 which is still not worth it for workstations.

If you are cloud only company then for sure it is just easier but still it wont be cheaper just more convenient to use. If data science team is very big probably "best" solution without unlimited money is just to run local and go cloud [premium] if you dont have free resources for your teams (It was the case when i was working in pretty big EU bank but it wasn't "true" Deep learning yet [about 4-5 years ago]).

Ahh yes cause there is only one way to do Deep Learning and it is ofc stacking models large enough to not be useful outside pods of GPUs and this is for sure way to go if you want to make money (from VC ofc cause you wont have much users that are ever willing to pay so much that you'll ever make even, as was OpenAI and other big model providers, maybe you can get some money/sponsoring from state or uni).

Market for local small and efficient models running on device is pretty big maybe even biggest that exist right now [ios, android and macos are pretty easy to monetize with low cost models that are useful]. I can assure you of that and you can do it on even 4x RTX 3090 [ it wont be fast but you'll get there :) ]

What cloud is even remotely worth it over buying 20x rtx 3090 or even some quadro for training? Maybe if u have very small team and small problems but if you have CV/Video tasks and team more than 3 maybe even 2 people in house servers are always better choice as you'll get your money back in 2-3 months of training over cloud solution and maybe even more if you wait for rtx 4090.

And if you are solo dev its even easier choice as you can reuse your rig for other stuff when you dont train anything (for example gaming :D).

Only possibility is if you get free 100k from AWS and then 100k from GCP you can live with that for a year or even two if u stack both providers but it is special case and im not sure how easy it is to get 100k right now.

Cold (i dont think estonia is cold but whatever) > warm in times of global warming when you dont even have enough water to support nuclear reactors in France.

South Europe (outside south France) is probably worst place to live if you care about 10+ years window and plan or have kids. You'll just let them out to live in 25% unemployment, high possibility of another immigration crisis and stagnant economy cause you prefere 40+ celsius over 25 in the middle of the day.

Is i do smth with DALL-E auto top hacker news post i saw like 20 post like that in past 2 weeks.

CoreML is such a bad lib compared to Pytorch/Jax and even Tensorflow that im not sure why anyone should use it if they can just use other stuff (i would probably prefer pure MPSGraph for training on device at it allows me to control everything in easy and Tensorflow like way)

+1 for fast.ai it is and was probably best course for noobs for years (i rember 2017/18 courses as big help in DL even that i got exp with machine learning before that [about 1 year of mostly classical ML] :P)

Its probably biggest environmental catastrophy in Europe after chernobyl, and the best of all polish goverment officials said on wednesday: "Everything is fine, you can go swim in the river and catch fish" and 2 days later they revert it all and send warnings to all people living near the river and this proceder was going on for at least 2 weeks.

I always prefer to use native > Desktop [qt] >>>> web apps.

They already try it like 2/3 times and it didnt take off so why should it be different today?

Still macOS language, a lot of cool features but not enough community on linux and for sure windows :/

Im not sure about waste of money. It is biggest US chip maker, you can for sure point on better chip reusing companies (NVIDIA and maybe AMD [they got a lot of good launches in past 4 years but it can change very quickly]). If USA want to be independent from China destroying/blocking Taiwan there is no other way then manufature your own chips (China and Taiwan put a lot of subsidized into lagging tech all the time and if it clicks it is always big just like TSMC)

Also Tesla grow on US gov support so don't fanboy 2much about that.

Swift is in the same or better performance bracket then C#/Java and unsafe usage can be close to performance of C (maybe 1.3-2x slower). I'm not sure about ranting for poor single test results :P

i dont have problems with OSX gui while KDE crash regulary. KDE can for sure looks better but it is not even close to same polish lvl as windows/osx

It is for sure not more polished then OSX or windows but it is best in Linux world if u care about looks not stability.

If you can buy consumer grade GPU i'll for sure prefer that as you can buy 4x rtx 3090 in same price as one A6000 and for almost all use cases outside extremely big models it should be a lot better solution. (I have one PC with 4x rtx 3090 for DL and it is enough but sometimes as i need to run very big model i just turn on colab with TPU [inference] or rent A100 server as it'll be still cheaper then any other solution)

For CPU I cant say much cause we are always limited by PCIE speed in our use case and Xeons are idle most of the time :P

how are u compiling (optimization, custom compilation flags etc.?) In my case https://github.com/mratsim/Arraymancer big project compile under your 4.2s so or you have like 10k+ lines of codes with macros or you just pass some debug flags to compiler :D

You are wrong for example Poland GDP per capita growth 6x Czech 3x Hungary 2x before they join EU and after commie ended (90-93) etc etc. https://data.worldbank.org/indicator/NY.GDP.PCAP.CD?location...

Yes thats why im showing post commie east europe as example of countires that was poor and are "rich" today

Im more about "East Asia is the only region which was certainly 'poor' but now can be described as somewhat 'rich',"

East/Central Europe (post commie)?

Ok show me how to use Rust for my big deep learning projects or even in big iOS projects and i'll switch :D

Citing other comment -> "Tachyum is writing press releases like a machine gun (2-3 a month), but they don't participate in MLPerf and avoid comparisons altogether. They talk about big facilities, but they have 43 employees (including ex-employees) on LinkedIn.

Doesn't anything confuse you?"