fnands

10 ms·

121.
by fnands 1y ago
Cool, but what's the Weissman Score?
122.
by fnands 1y ago
Fun to see we are now measuring GPUs in gigawatts
123.
by fnands 1y ago
Mojo might be what you are looking for: https://docs.modular.com/mojo/manual/gpu/intro-tutorial/ The language is general, but the current focus is really on programming GPUs.
124.
by fnands 1y ago
It's still very much in a beta stage, so a little bit hard to use yet. Mojo is effectively an internal tool that Modular have released publicly. I'd be surprised to see any serious adoption until a 1.0 state is reached. But as the
125.
by fnands 1y ago
> The strong AI focus seems to be a sign of the times, and not actually something that makes sense imo. It has been Mojo's explicit goal from the start. It has it's roots in the time that Chris Lattner spent at Google working o
126.
by fnands 1y ago
Anything to look out for for someone considering going for LASIK in the next few months?
127.
by fnands 1y ago
Their repo has some example notebooks: https://github.com/facebookresearch/dinov3/tree/main/noteboo... As for doing it in general, it's a fairly standard vision transformer so anything built on DINO
128.
by fnands 1y ago
As someone who works on satellite imagery, this part is incredibly exciting: > ViT models pretrained on satellite dataset (SAT-493M) DINOv2 had pretty poor out-of-the-box performance on satellite/aerial imagery, so it's super e
129.
by fnands 1y ago
Thanks!
130.
by fnands 1y ago
I am way OOTL with crypto drama. Anyone have any context about who Qubic are, and what their deal is?
131.
by fnands 1y ago
I had an Amazfit too a while back. Decent watch for the price, but battery life absolutely tanked after a year of use. Went from lasting days to lasting hours after one year.
132.
by fnands 1y ago
Also: > The user appears to think the model is "gpt-oss-120b", a new open source release by OpenAI. The user likely is misunderstanding: I'm ChatGPT, powered possibly by GPT-4 or GPT-4 Turbo as per OpenAI. In reality, ther
133.
by fnands 1y ago
Mhh, I wonder if these are distilled from GPT4-Turbo. I asked it some questions and it seems to think it is based on GPT4-Turbo: > Thus we need to answer "I (ChatGPT) am based on GPT-4 Turbo; number of parameters not disclosed; GPT-
134.
by fnands 1y ago
> Talking a PM down from a task that was never feasible One of our EMs did this this week. He did a lot of homework: spoke to quite a few experts and pretty soon realised this task was too hard for his team to ever accomplish, if it was
135.
by fnands 1y ago
Damn, I'm getting Black Mirror vibes from this. Maybe because I watched the Eulogy episode last night. Really great work though, impressive to see.
136.
by fnands 1y ago
So long and thanks for all the slurs? ;-)
137.
by fnands 1y ago
There's also a fish with the word "Balls" written on it: https://drawafish.com/profile.html?userId=1754400041983_0hqg... Pretty mild, all things considered.
138.
NASA-Isro Synthetic Aperture Radar (Nisar) Launched Successfully(bbc.com)
6 points by fnands 1y ago|0 comments
139.
by fnands 1y ago
I don't think they've found anyone on the big tours, but iirc this was the first case of someone getting caught: https://www.bbc.com/sport/cycling/36142963
140.
Translating Cython to Mojo, a first attempt(fnands.com)
2 points by fnands 1y ago|1 comments
141.
by fnands 1y ago
I kick the tires on Mojo’s new Python interop by translating some of scikit-learn’s Cython into Mojo
142.
by fnands 1y ago
Interesting! I built something similar at a hackathon a while back: https://fnands.com/blog/2024/factory-hackathon/ We called it Journalaist, and billed it as a personal ghostwriter. What we found is that it
143.
by fnands 1y ago
Have you tried Mojo? There all the numeric types are SIMD by default (if you declare a UInt8 it is actually a SIMD vector of UInt8s of length 1 by default). https://docs.modular.com/mojo/manual/types/#simd-and
144.
by fnands 1y ago
Working with elevation data in the US is a pleasure compared to most other countries. Freely available Lidar and derived products (via the USGS API) for nearly the entire country makes working with elevation data a pleasure.
145.
by fnands 1y ago
Cool, makes sense. Yeah, have you considered maybe looking into just running it on embeddings [1], instead of the imagery itself? Would save on most of the inference cost, at the cost of flexibility (i.e. you are locked into whatever embed
146.
by fnands 1y ago
Hey, cool project! Do you actually pass the images to the model, or just the metadata/stats?
147.
by fnands 1y ago
They do have fibre, but a large part of (non-dried) mangos are water. If you remove the water, then the sugar density per volume/weight goes up a lot. This makes it easier to consume much larger volumes of dried mangos vs raw mangos.
148.
by fnands 1y ago
> hundreds of interested and highly capable people to agree That's a polite way of putting it, but 100% the grit required to push this through is admirable. > Of course he'll be on a lot of papers now :-) Well, everyone in A
149.
by fnands 1y ago
> six-year quest to convince the collaboration of the value of the new technique The technical achievement is really impressive, but as someone who was on ATLAS this is the most impressive part. The amount of work it takes to convince th
150.
by fnands 1y ago
GPU programming in Rust isn't great. In Mojo it's pretty much the whole point of the language. If you're only using CPUs, then yeah, PyO3 is a good choice.

More ›