+1. There are multiple comparisons in the paper, and we plan to provide a few more soon. The typical story that if the model is at least a bit unusual, not overly optimized manually, Hidet can provide some speedup even over TensorRT.
If you read into the paper (https://dl.acm.org/doi/10.1145/3575693.3575702), one can find more performance comparisons.
There, from a latency/throughput PoV they are en par with existing tools like TVM/Ansor. Sometimes faster, sometimes slower.
What is more interesting is this: They have very GPU-specific auto-tuning routine that drastically reduces the optimzation space, compared to TVM/Ansor. They go from ~10^6 possible implementations for an operator to a "few hundred", which enabled much faster time-to-solution. This is achieved with a GPU-centric problem formulation and search space. In essence, they trade how widely applicable their approach is (from "any" kind of hardware to only GPU-style architectures) for retrieval speed.
Im currently working on a project, involving a language I‘m not familiar with.
It is super simple to paste a function or section of code, have the functionality explained to me and iterate on any open questions.
Last year, a Buell XB12R Firebolt. An exotic, American sport bike with a radical design and engineering philosophy. I knew they existed, but never paid much attention. Then a distant friend sold his bike. I took a test ride, just for the fun of it, because they are quite rare.
On the ride home I already had made the downpayment with the intent to pick up the bike the next day. One the best decisions I ever made!
This reminds me of a movie, the name of which I can’t remember.
> A family father gets premonitions about a massive storm that will come, which triggers him building a bunker in his back yard. It is a family drama on how his obsession with protecting his family slowly tears it apart. In the end they convince the father to go on vacation for a few days, just to get away from the obsession. As they arrive in their vacation home, a huge, black storm cloud rolls in on the horizon
The question is, do we really need a PhD reform? Or should we restructure the academic process to enable more people to pariticipate without going through the ardous process of writing a thesis? I have a PhD and work in an industrial research environment, with 50/50 PhD and non-PhD colleagues. There is no difference in quality of work and output. So maybe universities should be less discriminatory against non-PhD research fellows.
> I have a PhD and work in an industrial research environment, with 50/50 PhD and non-PhD colleagues. There is no difference in quality of work and output. So maybe universities should be less discriminatory against non-PhD research fellows.
I don't disagree. But if we consider that researchers need to publish papers, requiring a PhD isn't not much different than requiring a bunch of publications (when you have 3-4 publications, basically you have a PhD thesis, just need to introduce context and glue everything together).
To be fair, this is really field dependent. In the sciences this is (mostly) true, but in the humanities your output for your PhD is sometimes just your thesis. The problem is fundamentally that the PhD traject is geared towards an academic career, but there are not enough academic positions for all those PhD students, so they end up in industry. And there their skills don't really translate that well, as OP also says.
>The problem is fundamentally that the PhD traject is geared towards an academic career, but there are not enough academic positions for all those PhD students, so they end up in industry. And there their skills don't really translate that well, as OP also says.
This is a gigantic problem and thank you for highlighting it. Imagine you're in the United States and you have a PhD but you're not a citizen here and you just invested a ton of time and effort but you have to go back home all because there's not a position in your field. It's absolutely nonsensical to think about how the United States justs pisses away talent.
yet it is used to gatekeep positions. Contributions to science can be made without bundling them in a thesis, while being delivered to the whims of your supervisor.
Then one should make a good case for changing that gatekeeping, if you think it’s unhelpful. That’s not what this article advocates: this article is about changing the PhD process inside the academic system.
Because there is an inevitable translation step between "customer wants X" and "engineers have to build Y to achieve X". This takes time and is a non-trivial task. I'm an engineer myself and I'm glad there are people between me and the customers. They're there to shield engineers from unnecessary tasks and scope creep. They are the necessary abstraction layer between engineering and business.
