JonChesterfield

8 ms·

61.
by JonChesterfield 10mo ago
Corrupts data on power loss according to their own docs. Like what you get outside of data centers. Not reliable then.
62.
by JonChesterfield 10mo ago
Per machine. Definitely more than one machine here.
63.
by JonChesterfield 10mo ago
Microsoft pleading poverty doesn't really fly
64.
by JonChesterfield 10mo ago
> The docs I upload are ones I'd be OK getting leaked. That also includes code. That's fortunate as uploading them to a LLM was you leaking them.
65.
by JonChesterfield 10mo ago
Ah right, in the nim repo, not the authors one. Contains https://github.com/nim-works/loony/blob/main/papers/GierschE... indeed, thank you
66.
by JonChesterfield 10mo ago
Mozilla spend a lot of time telling me I trust them. I don't think that's having the effect they expect.
67.
by JonChesterfield 10mo ago
The paper is annoyingly difficult to locate but the author's implementation is at https://github.com/oliver-giersch/looqueue-rs
68.
by JonChesterfield 10mo ago
I should check this next time I take it apart, thanks for the heads up :)
69.
by JonChesterfield 10mo ago
Boiling at lower power uses _more_ energy to reach the same temperature. This makes a 2kW kettle on eco grounds especially dumb, yes, but that doesn't preclude people pushing for it.
70.
by JonChesterfield 10mo ago
I'm sure there was a period where one couldn't find a 3kW kettle in the UK on power efficiency grounds, one was supposed to run a 2kW one instead to save the planet. But now when I search I find 3kW models again. So either that wa
71.
by JonChesterfield 10mo ago
Uh shite. That's the goto for ssds with capacitors on, such that when the power dies they're able to leave things in a reasonable state. All the samsung ones I've tried have died within a year. Fast until they died I suppose.
72.
by JonChesterfield 10mo ago
The strix halo mini workstation is 128gb of ecc. Good machine. Source https://h20195.www2.hp.com/v2/GetPDF.aspx/c09086887
73.
by JonChesterfield 10mo ago
We've written a cuda to amdgpu compiler. It should be possible to persuade someone like Modal that this is wonderful. So far people largely decline to believe that it can be done at all :) If you've written vaguely reasonable cuda
74.
by JonChesterfield 10mo ago
Well, SO is only read by scrapers these days so no answers doesn't mean much. Could be the question is unsolvable, could be lots of people know the answer and won't see it because you've written it in a graveyard. edit: thoug
75.
by JonChesterfield 10mo ago
Usually temporal logic of actions. Lamport's thing. Extremely worth a look.
76.
by JonChesterfield 10mo ago
There's a coherent description of what the product is at https://www.producthunt.com/products/gitarsenal?comment=5026... - the linked website is too fancy for me to make sense of. Can't throw stones though, o
77.
by JonChesterfield 11mo ago
The content will be discovered just fine. It'll get embedded in the LLMs on the next round of training. Won't be attributed to your blog of course, but an approximation to the information will still get out there.
78.
by JonChesterfield 11mo ago
I take that back, returning their laptop after two days.
79.
by JonChesterfield 11mo ago
HP are your friend. In laptop or desktop form factor, 128gb ram.
80.
by JonChesterfield 11mo ago
What's the local workaround for this? Git is distributed, it should be possible to put something between our servers and github which pulls from github when it's running and otherwise serves whatever it used to have. A cache of so
81.
by JonChesterfield 11mo ago
Couldn't log into it this morning when cloudflare was down so there's that.
82.
by JonChesterfield 11mo ago
The chip in general is fast, it builds llvm in ~12m or so. Whisper on it is at least real time but I only ran the stream binary before sending the box away to SC25. I'm expecting it to need some work to exploit the zero copy the APU pe
83.
by JonChesterfield 11mo ago
Guarding builtin_expect with __GNUC__ instead of has_builtin might be what you find in elderly codebases that were fed into an llm
84.
by JonChesterfield 11mo ago
shouldn't be individually allocated either, arenas for the win
85.
by JonChesterfield 11mo ago
I too enjoy space leaks on recursion
86.
by JonChesterfield 11mo ago
I got one of those running whisper yesterday, hopeful the bigger llms will run shortly. You'd need rocm 7 which seems to be much better than 6.4 was.
87.
by JonChesterfield 11mo ago
Anyone know whether there are things built on https://github.com/HazyResearch/ThunderKittens ? I think this is a port of that to HIP, where generally ports of cuda things to hip are of vague professional interest, but m
88.
by JonChesterfield 11mo ago
This doesn't sound right. I definitely got yelled at over trivial performance regressions which looked like noise so people were measuring performance. They've paid serious amounts in RSUs over the last six years. Not top of marke
89.
by JonChesterfield 11mo ago
That one is historically interesting. I suspect the first compilers were named that because they were making compilations of assembly routines, probably slightly modified/specialised to the rest of the routines. Compilers still do that
90.
by JonChesterfield 11mo ago
Speculative development would be a problem in the quarterly account filing. Good idea on longer horizons yes.

More ›