Job

You tune the column prompt on it, read the outputs, fix the prompt, and only then go big.

For bigger runs, the write-up has you take that exported config and run it as a job on Hugging Face Jobs.

I like that, because the config is the bit you want to keep.

The catch is that Jobs builds by the minute while it's starting or running, and what's documented is mostly command line and API, not a screen.

So the design is a GUI, and the scale is a terminal.

Pretty much the trade.

And then there's the layer most teams miss entirely.

Jobs, caches, and queues.

If you need it tonight, you pick 12.

And the gotcha? Jobs on unsupported models can sit pending without throwing an error.

You'll be waiting for nothing and no one will tell you.

Grr.

So for batch

work, jobs beats endpoints.

Jobs beats endpoints by a lot.

Endpoints are for serving traffic.

And what did fixing it actually save?

After right-sizing the targeted jobs based on that data, weekly compute costs dropped 31%, measured across a four-week period before and after the change.

Jobs also started running more than 7% more often in that same window.

Did performance get worse anywhere to pay for that?

Not based on what they reported.

A

few months before this, Steve Jobs and Bill Gates held a conference and announced this release, and then the software came on September 30th.

It's interesting why Excel came out first.

There were a number of reasons why that happened.

The synergy didn't seem to be there in terms of development.

I don't know that for sure, but it just seemed like that would be something that Microsoft would time for the release of Windows, but they didn't.

Steve Jobs probably worked on some sort of partnership deal with exclusivity, not necessarily exclusivity, but gave Microsoft an incentive, so to speak, to release it first on the Mac.

this was um the same year in january of this year they they tried to do macintosh office that was that super bowl ad