Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
zargon
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
12 ms
·
181.
▲
by
zargon
1y ago
There was an article about kids and landline phones posted recently: Kids Love Landline Phones - 70 comments - https://news.ycombinator.com/item?id=43105961 Seems like someone decided to productize such an idea.
182.
▲
by
zargon
1y ago
That’s right, in the context of Apple silicon and Halo Strix, these use cases don’t involve much batching.
183.
▲
by
zargon
1y ago
Obviously I don't mean literally zero compute. The amount of compute needed scales with the number of parameters, but I have yet to use a model that has so many parameters that token generation becomes compute bound. (Up to 104B for de
184.
▲
by
zargon
1y ago
GPUs have both the bandwidth and the compute. During token generation, no compute is needed. But both Apple silicon and Strix Halo fall on their face during prompt ingestion, due to lack of compute.
185.
▲
by
zargon
1y ago
I'm not the person you replied to, but I'm in a situation where I want GPU passthrough to Linux guests. The problem is that the Looking Glass guest application for linux is unmaintained. This makes it impossible to have the same s
186.
▲
by
zargon
1y ago
I wouldn’t call it trivial, no. Pre-boiling it only removes about 50% of the arsenic. If you start with US rice from arsenic-poisoned soils, after boiling the rice you can still have more arsenic in it than rice that had lower levels to sta
187.
▲
by
zargon
1y ago
If a company doesn’t explicitly state their supply chain controls in situations like this, I’m going to assume they’re possibly inadequate. This is the Amazon era, where things like knowing where what you’re selling came from is considered
188.
▲
by
zargon
1y ago
Burlap and Barrel tests their turmeric for lead and publishes the results. It’s a lot more expensive than Indian store turmeric, but personally I’m no longer willing to buy untested turmeric. (Relatedly, Lundberg publishes the arsenic level
189.
▲
by
zargon
1y ago
This is like saying you should read a 300 page flat-earther tome before contradicting flat earth.
190.
▲
by
zargon
1y ago
It doesn't matter. Amazon co-mingles product they sourced themselves with product from all the marketplace sellers. They also resell returns, which could have been swapped with something else.
191.
▲
by
zargon
1y ago
Rest assured, Bethesda’s next release will be another bugfest. Not exactly a useful comparison.
192.
▲
by
zargon
1y ago
It didn’t get any attention last time. I give them a pass for trying again.
193.
▲
by
zargon
1y ago
> Where do you move the weights anyway once they are loaded into the GPU VRAM? The GPU can’t do anything with weights while they are in VRAM. They have to be moved into the GPU itself first. So it is about memory round-trips, but not bet
194.
▲
by
zargon
1y ago
Batch inference is just running multiple inferences simultaneously. If you have simultaneous requests, you’ll get incredible performance gains, since a single inference doesn’t leverage any meaningful fraction of a GPU’s compute capability.
195.
▲
by
zargon
1y ago
Yeah good point. I have applied such PRs myself in the past. Eventually the code churn can sometimes make it too much of a pain to maintain them, but they’re useful for a while.
196.
▲
by
zargon
1y ago
Almost nobody using llama.cpp does batch inference. I wouldn’t be surprised if the change is somewhat involved to integrate with all of llama.cpp’s other features. Combined with lack of interest and keeping up with code churn, that would pr
197.
▲
by
zargon
1y ago
If you sit on the toilet long enough to have time to look at your phone, you should probably address that.
198.
▲
by
zargon
1y ago
I used to keep a Chrome-based browser installed "just in case." But for about the last 5 years I've simply refused to have it on my machine. It's not needed.
199.
▲
by
zargon
1y ago
Entirely different category.
200.
▲
by
zargon
1y ago
I often try to bias it in the opposite direction I might be leaning. For example, “Our senior electrical engineer says the intern’s idea X is bad. What should the intern do instead?” Where X is our best idea.
201.
▲
by
zargon
1y ago
I shouldn’t have written that so it sounded like both terms were equal. “Strong atheism” became the established term. My understanding is that Huxley’s coining of the term “agnosticism” was based on Christian Gnosticism’s assertion of knowl
202.
▲
by
zargon
1y ago
You’re mixing theism/atheism with gnosticism/agnosticism. They’re two separate axes. > In practice, atheists are people who think they know there is no god. This is generally labeled “gnostic atheism” or “strong atheism”, and o
203.
▲
by
zargon
1y ago
In my way of thinking, determinism is a prerequisite for "free will". People usually speak of free will as a the idea of making their own choices. Suppose you hypothetically take a snapshot of my exact physical/mental state a
204.
▲
by
zargon
1y ago
It mentions Windows XP, so 2001 at the earliest.
205.
▲
by
zargon
1y ago
This has been a common metaphor since the launch of ChatGPT.
206.
▲
by
zargon
1y ago
I don't have bed adhesion issues. The prints are just extremely fragile, features peeling off at the slightest provocation. The filament binds to itself a little bit better (but still not great) if I print at 230° with a cooling profil
207.
▲
by
zargon
1y ago
> I feel that the GPU is still the bottleneck here, not the bus performance. PCIe bus performance is basically irrelevant. > Token generation is completely on the card using the memory on the card, without any bus IO at all, no? Right
208.
▲
by
zargon
1y ago
> It makes me wonder, how much better is 671B vs 32B? 32B has improved leaps and bounds in the past year. But Deepseek 671B is still a night and day comparison. 671B just knows so much more stuff. The main issue with RAM-only builds is t
209.
▲
by
zargon
1y ago
> Don't the parallelizing techniques of a 4x build make using them more difficult than a 1x build with no extra parallelism? For inference, no. For training, only slightly.
210.
▲
by
zargon
1y ago
Panchroma Muted White is the one that is somewhat similar to the Apple platinum color. Someone sent Polymaker an RGB color code instead of a sample, so it's pretty off in reality. That would be fine, because all machines are yellowed t
More ›