luyu_wu

It is wild that this a log scale of cost to me!

For those who didn't check the page yet, it just links to the API docs being updated with the upcoming models, not the actual model release.

The iGP can access shared memory the same as with any modern system. I'm confused where you got this idea from.

Keep in mind the Ultra 300 chips also only have recent support in the kernel. The battery life likely isn't great for now (as with previous gen Intels right after release). It makes sense to me that for now the benchmarks would be Windows specific.

Consider reading the article, which addresses all of the points you raise.

It's directly stated in the post that the entire test is meant to be humorous, not taken seriously, only that is has vaguely followed model performance to date. The author also writes that this new result shows that trend has broken..

This article seems overly critical trying to impose a stance. I have never heard anyone say "因为雨下得很大，所以我决定不去了".

> The Sausage Sentence: English stacks relative clauses. Modern Chinese attempts to shove that complexity into a single pre-noun modifier using de (的), creating bloated, breathless sentences that tax the memory.

This is given without any evidence. "Creating bloated, breathless sentences that tax the memory" sounds like something Claude might write. IMO, 的 is far from as negative as the author (or AI) portrays it; arguably better than the multitude of English synonyms (his, her, theirs, its).

It's likely you are pulling one to ground somehow. That was a common bug I faced.

If I remember correctly,it was taped out by some company as some embedded core in a GPU?

I guess that may be the true use case for 'Open-Source' cores.

That being said, the advertised SPEC2007 scores are close to a M1 in IPC.

I think this is quite interesting for local AI applications. As this technology basically scales with parameter size, if there could be some ASIC for a QWen 0.5B or Google 0.3B model thrown onto a laptop motherboard it'd be very interesting.

Obviously not for any hard applications, but for significantly better autocorrect, local next word predictions, file indexing (tagging I suppose).

The efficiency of such a small model should theoretically be great!