How do you get data loss?keepassXC,DX saves a conflict copy and warns you. Anytime I've seen the warning over ~10 years it's been a non issue. Like I add an entry on PC, walk away from the 'save db' prompt for a day and then update something on my phone so I have 1 new account on both. I see the warning and so I have to hit one button to do the basic merge or whatever and it's done.
That's what I'm saying, a lot of people are coping with a product they admit will need a fork.
Not only is it incurring the cost of project fragmentation, but also incurring an always online cost with overly-complicated docker solutions, when a fully offline and airgapped solution already exists.
Furthermore, staying with the same ecosystem invokes the sunken cost fallacy. But the migration from Bitwarden couldn't be simpler (just export Bitwarden json file). It's almost a form of battered woman syndrome people are inflicting on themselves when quite simply they can hop onto an already proven ecosystem that doesn't bait and switch.
I was on keepass before bitwarden. Bitwarden just solves more things for me. I am sure the keepass ecosystem improved a lot over the years but fundamentally i find vaultwarden docker to be far easier. Especially for my work and family members that i convinced to use bitwarden. If they were also in charge of the sync it wouldn't be possible.
Afaik vaultwarden and bitwarden clients are as proven as keepass.
I got a TI83 in 4th grade and I realized programming is how you made video games and I decided "ok I'm going to learn to program"
I read the whole manual's programming section but couldn't make heads or tails of it. It assumed you knew basic logic/programming and mostly explained functionality/syntax.
Then in 5th grade my friend who was 3 years older was like "hey look I made a story in my calculator" and it was this big choose your own adventure story. He showed me how to use goto, how to display text, and a function for multiple choice user input + goto. I was in business!
I wrote my own story but had a section where I wanted to do different things if you had gotten an item already so I had to program the whole story twice and only enter the second version from the option where you get the item.
I tried writing a more complicated story with more items but the duplication was insane 3 items required 3!=6 copies of common locations. I was like "this is dumb there's got to be a better way" and I looked at the manual again and now I had enough of a framework to understand "OH a variable is whether you have the red key, why didn't they just put that?"
the industry has largely moved away from QAT because the hardware required for running a quantized model are an order of magnitude less than training/QATing the fp model.
That's why things like Autoround, GPTQ, AWQ have been so popular, you don't even need enough hardware to run the original model on gpu, just cpu is enough due to the data efficiency
Thanks. I think it is a good explanation, but also suggests a gap. QAT to me, if done right, is the only way to recover performance for extreme quantization regime. The only thing matters of course, if whether it can work. My confidence in QAT comes from the LoRA can recover most quality misses in quantization, and that is still different from QAT for extreme quantization, so it could be very wrong. I need to try it anyway.
the Turboquant debaucle is entirely on Google doing incredible amount of academic misconduct and general incompetence.
A) the QLJ thing they added is useless and the code they released didn't even include it since it makes the results worse
B) the blogpost about turboquant was ai generated and stated that turboquant used a polar transformation when it doesn't, so for the first 2 weeks people thought turboquant involved a transformation to polar coordinates. The reason the blogpost was wrong was probably because google tried to put their useless polarquant paper on the map by talking about it repeatedly.
C) since they don't use QLJ they wholesale copied the quantization technique from the HIGGS paper without citing it (https://arxiv.org/pdf/2411.17525)
What are you guys doing to get real issues?