namibj

36 ms·

691.
by namibj 3y ago
Yeah the shielding is to not cook your eyeballs if you observe the food, and to not cook anything else outside of the chamber, including e.g. your phone's wifi receiver.
692.
by namibj 3y ago
AFAIK it's the other side of the die from a "delidded" view. So, they desoldered the die, then etched/ground at it until the surface coating was gone, and just the semiconductor structure remains. Due to the sub-micromet
693.
by namibj 3y ago
Funnily that's applicable to balancing unswept flying wings, as all high lift/low drag airfoils are unstable in pitch (though many are stable for inverse flight, i.e. if you get scared you'd have to turn upside down and are s
694.
by namibj 3y ago
If you want efficiency you'd use iirc 405nm laser diode plus blue LED receiver.
695.
by namibj 3y ago
Well, you still don't get fine-grained random deletions on Zoned NVMe. Also, there is another storage target which expects append-only style: SMR HDDs. They don't support random writes into a zone.
696.
by namibj 3y ago
You could quite easily check if anything you care about was touched by the previous lock holder, so that they could indeed write everything but the root in parallel/concurrently. E.g. note batched operations, like "insert list ele
697.
by namibj 3y ago
Tell that to NVMe zoned namespaces. Outside of RAM, immutable data structures exist for device technology (e.g. MLC NAND) inherent reasons.
698.
by namibj 3y ago
We could do computer-only search for enzymes that catalyze desired reactions in enzyme-friendly reaction environments. Assuming that the way that makes the above 100% accurate also yields us the tooling to simulate the candidates in operati
699.
by namibj 3y ago
And German: Erdgas.
700.
by namibj 3y ago
> Current performant (high accuracy) LLMs have a quadratic cost (space and time) in sequence length, FlashAttention only takes polylogarithmic space overhead beyond the linear needed for attention. Yes, quadratic compute, but that's
701.
by namibj 3y ago
Materializing the attention matrix has you slower than FlashAttention much of the time, just because you get severely memory bandwidth bound and are worse off than mild extra computations resulting from streaming it (and especially re-compu
702.
by namibj 3y ago
The security nightmares talk at (also) 37c3?
703.
by namibj 3y ago
You can easily use pipeline-parallelism though. Especially if you have 8-16 lanes of PCIe4 with direct P2P access between the cards. IIRC you want micro-batching though, to overlap pipeline phases.
704.
by namibj 3y ago
Critically, they usually involve forbidden intermediates in an approximation of Feyman diagrams where the degree of the graph with quantum interaction nodes/vertices between particles/edges is bounded. The way out is contracting t
705.
by namibj 3y ago
No, an active converter from a PCIe power connector or such to SATA power. Perhaps integrated in many sata cards.
706.
by namibj 3y ago
Transformers don't need quadratic memory for attention unless you scale the head dimension proportional to the sequence length. And even that can be tamed. The arithmetic intensity of unfused attention is too low on usual GPUs; it&#x27
707.
by namibj 3y ago
In Germany regular grocery stores are in walking distance (if your living rural, you might strongly prefer a bicycle, though), and last I looked, the mandatory post-checkout secondary-packaging-disposal/-recycling station includes a ba
708.
by namibj 3y ago
The issue in that respect is that they flood the coils in a helium bath, instead of utilizing additional structures like channels/nozzles to provide the same rate of freshly cooled helium to the heat sources as the bath does, but with
709.
by namibj 3y ago
Replace the x with a 5 in the abstract URL.
710.
by namibj 3y ago
The older generations had substantial docs; they were iirc basically MPLS on their L2. With a bunch of e.g. remote atomic verbs that even worked through to POWER9 hosts, great for throughout on fetch_and_add heavy (almost) wait-free data st
711.
by namibj 3y ago
I could imagine transmissive at the belly button to work out with decent design effort, or maybe around the ear. Nostrils seem unlikely to be comfortable, but would probably work fine.
712.
by namibj 3y ago
Also it's basically an Mi50 in a trench coat. You could buy a 32 GiB version labeled Mi60, initially with a single mini DP port making it IMO classify as "graphics card" not "compute accelerator".
713.
by namibj 3y ago
I had FedEx send me to collections for duties on a parcel marked to not deliver without collecting duties, for which the delivery guy signed a receipt. Horrible.
714.
by namibj 3y ago
Yeah, real world is messy. Also, the contribution/influence of ancient state in softmax is something the controller can learn, especially with a task-suitable position encoding. Though I'd not be surprised if what's IIUC call
715.
by namibj 3y ago
Yeah, those helmets that are basically face shields with optional hard hat and neck hood are generally speaking fairly comfortable: wide FOV, zero difficulty breathing, and an entire face shield (not just eye "shield") almost casu
716.
by namibj 3y ago
The feed forward layer is essentially a differentiable key-value store. Similar to the attention layer, actually. So it just uses an attention mechanism like pre-selector to attend to only some experts. During inference, this cutoff is made
717.
by namibj 3y ago
Actually somewhat better, even: you can let it control the rate factor of your video encoder directly, getting perceptual fairness instead of simple naive bandwidth fairness.
718.
by namibj 3y ago
The math works (I think, don't really see why it shouldn't); the intermediate of the scan is matrix-valued, though.
719.
by namibj 3y ago
They are ((sub-)orbital) rockets, not atmospheric bullets. I'm not surprised it was hard to compute to the accuracies desired.
720.
by namibj 3y ago
Ehhh, video core is basically just a (iirc multi-core/-thread) vector processor. Funnily enough, this also makes these rather cheap number crunching hard real time chips with the high-bandwidth IO (for hard-real-time) of a Pi. Notably,

More ›