flounder3

Relevant PR: https://github.com/ggml-org/llama.cpp/pull/27742

This branch works now: https://github.com/unslothai/llama.cpp/tree/qwen4exp/qwen3.8...

  cmake -B build -DGGML_CUDA=ON

or

  cmake -B build -DGGML_METAL=ON

then

  cmake --build build --config Release -j --target llama-server llama-cli

One traditional enterprise goal of 40% utilization was to cover DR/failovers, so one region could take on 100% of traffic from another, with 20% headroom.

I'm curious about the granularity of contracts around granting/selling excess capacity. Are they short term? Can the owner evict those workloads (with a penalty)?

WiFi. Flip it on for an update, then leave it off.

> do dealers have any way to update control modules besides OTA?

Yes.

Muscle memory for folks who have been doing it since before -i was an option. I still instinctively type `sudo su -` because it worked consistently on older deployments. When you have to operate a fleet of varying ages and distributions, you tend to quickly learn [if only out of frustration] what works everywhere vs only on the newer stuff.

`sudo su - <user>` also seems easier for me to type than `sudo -i -u <user>`

> Without any concerns about confidentiality or respect for the persons family.

The link is to a recent post from the person's father, asking for awareness around the company's actions.

If you really are a retired psychotherapist, you should know and do better than to ignorantly call someone a "fucking ghoul," especially in light of your lack of awareness or willingness to educate yourself as to the context of the post.

Shame on you.

How many hours have you spent with the latest Tesla FSD in congested neighborhoods or unprotected left turns at sunset? It may be "good," but it's not even remotely close to the current Waymo experience, especially when it comes to abnormal situations. I am continually blown away by how Waymo behaves (spent 3 hours in them this past week). I'm rooting for Tesla, but it is nowhere near Waymo as of today.

OP's 9to5mac article states:

  Also under the agreement, Google’s model will reportedly run on Apple’s own servers, which in practice means that no user data will be shared with Google. Instead, they won’t leave Apple’s Private Cloud Compute structure.

Bloomberg states:

  The model will run on Apple’s own Private Cloud Compute servers, ensuring that user data remains walled off from Google’s infrastructure.

You must have worked in a very odd and isolated department. I never heard that rhetoric, even once, throughout my tenure. Nor have any of my old colleagues who still work there and are quite well known internally (notorious patents, features / tentpole DRI, etc).

Either a troll or somehow didn't get the memo that TSMC building OUTSIDE of Taiwan is a big deal these days: https://en.wikipedia.org/wiki/TSMC#Arizona https://en.wikipedia.org/wiki/TSMC#Washington https://en.wikipedia.org/wiki/TSMC#Japan https://en.wikipedia.org/wiki/TSMC#Germany

A recall means “PAY ATTENTION! Something may still be broken!”

The recall will mention how to get it fixed, regardless of a OTA update or service visit.

The lack of a mere download could mean the difference between life and death.

It’s the law. OTA updates are not instant nor are they 100% guaranteed, so consumers must be notified.

Was about to say the same thing. It was already blazingly fast, with a typical album only taking seconds.

I’m a backer and can’t wait to receive the 8TB upgrade!

I had to buy an off the shelf M1 Studio due to a hardware failure, so I couldn’t wait for the lead time for one with more RAM and storage. It has been borderline unusable due to so many things requiring local storage — can’t even symlink to an external NVMe. (Many apps, but also Backblaze metadata and iMessage attachments)

This is a drop in the bucket for photographers, videographers, and general backups of RAW / high resolution videos from mobile devices. 80TB [usable] was "just enough" for my household in 2016.

This is not news, and has been the case since Siri's inception.

On the very first page:

The internal hard drive uses SCSI with an unusual connector. Adapting it didn't seem straightforward, and we weren't confident the old file system (HFS) would be easy to read from a modern system.

PG&E in California is 4-6x+. I care.

https://www.pge.com/assets/pge/docs/account/rate-plans/resid...

We found that most proxies/firewalls (90%+ ? I forget) didn't tamper with it. The largest hurdle was working with load balancer vendors to implement it.

NUCs are a tremendous value over previous generation surplus hardware in areas where the cost of electricity has skyrocketed (SF Bay Area).

We’re talking more than one order of magnitude less power consumption when you compare to 2-3+ generations ago.

"Shit" or not, it was a fairly popular place to hang out while waiting for a train at Dirdon Station or an event at SAP Center. It was only a few steps away, and now there's nothing there (after existing for 88 years).

If you Google "San Jose Google Construction," you don't even have to click into an article to learn that there has already been a large amount of demolition.

You've clearly never run it at scale nor have you migrated between Enterprise (on-prem) and Splunk Cloud at scale. Managing .conf files and eliminating intermediate IDM logic was absolutely not "amazing."

https://lantern.splunk.com/Splunk_Platform/Splunk_Cloud_Plat...

You’re not wrong, but privacy concerns were paramount.

ASNs were allowed but too vague. We needed more granularity. Corporate proxies, subdelegations, many providers aggregating announcements below /24, etc.

Even 10 years ago, Apple internal privacy policies prevented itself from collecting precise lat/long. We had to use HTTP session telemetry to determine which endpoints were best for a given IP (or subnet, but not ASN), which informed our own pseudo-geoIP database so we knew which endpoint to connect to based on real world conditions.

Even still, it had to be as ephemeral as possible for the sake of privacy. We weren’t allowed to use or record results from Apple Maps’ reverse geo service outside of the context of a live user request (finding nearby restaurants, etc).

Chargepoint requires you to maintain an account balance, and requires auto-refills. You can't just bill a CC for the entirety of a session through the app/site. Sometimes you can initiate a non-linked session with an EMV tap from a physical card, but that doesn't always work.

No, it is not bullshit. And no, he certainly was not excellent. Perhaps he managed to take credit for a few "good" things, but it's interesting to watch you defend someone who was ~so~ widely hated.

I could give you a list of 40 tenured engineers and leaders in Eddy's org [along with at least 20 in adjacent orgs] who physically celebrated his departure.

This place is amazing for hosting events. They have a side room that’s perfect for a birthday party.

At the moment, it’s the third tweet down on the Replies tab on his profile.

> The only function of the box is to switch the power on and off

That is false. These are much more sophisticated than a mere relay. Sure, there is basic logic between vehicles and EVSEs which is basically a smart relay: https://en.wikipedia.org/wiki/SAE_J1772#Signaling

...but the electronics you mentioned are for onboard controls (energy monitoring, Time of Use enforcement, load balancing between multiple units, etc). Some EVSEs are very simple, but many -- including the one in the article -- also handle upstream communication like OCPP: https://en.wikipedia.org/wiki/Open_Charge_Point_Protocol https://www.chargepoint.com/files/OCPP-Fact-Sheet.pdf

Here is an example of integrations for the model the article is about: https://zaptec.com/downloads/ZapChargerPro_Integration.pdf

There are also implementations of power line communication over the Control Pilot pin in J1772, which most people notice when charging via CCS at a DC Fast Charger, but that is not limited to L3 charging. See also: Plug and Charge https://en.wikipedia.org/wiki/ISO_15118