ipsum2

24 ms·

481.
by ipsum2 2y ago
WSJ is wrong. Autocomplete code models are small (5B param) and very cheap. A single inference is roughly $0.000001.
482.
by ipsum2 2y ago
This is purely a scheduler, it doesn't have any any effect on the workload performance.
483.
by ipsum2 2y ago
Not quite. The closest equivalent would be something like fairseq. It's config (yaml) driven.
484.
by ipsum2 2y ago
It's a layer on top of PyTorch, and it has code to translate PyTorch models into MLX.
485.
by ipsum2 2y ago
This is a big deal for increasing performance past fp8. We'll see Nvidia/AMD/Intel etc adopt this in their products in the next year.
486.
by ipsum2 2y ago
It's interesting that Apple also actively develops https://github.com/apple/axlearn , which is a library on top of Jax. Seems like half the ML teams at Apple use PyTorch, and the other half uses Jax. Maybe its spli
487.
by ipsum2 2y ago
"If I've managed to build my own model, say a fine-tuned version of Llama"
488.
by ipsum2 2y ago
He mentioned he already has a fine-tuned model. OpenAI does not host user-created models.
489.
by ipsum2 2y ago
I implemented my own DIY version of Rabbit at a hackathon using Playwright and VNC. I feel extremely validated that they use the same things that I thought of.
490.
by ipsum2 3y ago
Curious about the story behind applying to YC as a founder and ending up as a moderator for hacker news.
491.
by ipsum2 3y ago
Nah, it meets the colloquial definition of open source, but not the strict OSI definition.
492.
by ipsum2 3y ago
Llama3 allows commercial redistribution and modification of the source code though.
493.
by ipsum2 3y ago
Meets the definition of the Wikipedia article you linked: "Open source is source code that is made freely available for possible modification and redistribution." Though, it doesn't meet the OSI definition.
494.
by ipsum2 3y ago
Is the passcode 02563? I got it twice, but its ignoring me when I ask if its correct. Is there something else that's supposed to happen?
495.
by ipsum2 3y ago
Thought it would be written in python with the name, but it's purely JavaScript.
496.
by ipsum2 3y ago
No, completely unrelated company.
497.
by ipsum2 3y ago
To a rough approximation, yes. The blackwell chip is also ~10x larger in surface area than MTIA, so the costs are proportional.
498.
by ipsum2 3y ago
That's unrelated to the article. If the article was about "how to build SpaceX rockets" the equivalent comment could be "I bought some fertilizer and made my own rocket using PVC pipes".
499.
by ipsum2 3y ago
It won't be under $10k.
500.
by ipsum2 3y ago
Airpods Pro v2 was released in 2022/2023 also uses AAC.
501.
by ipsum2 3y ago
I'd use them. I got my Airpods Pro v1 replaced once already because of the battery life. The v2 has no significant benefits to shell out $250 after only using the v1s for 2 years.
502.
by ipsum2 3y ago
It's a neural network, not a traditional compression algorithm. It would be difficult to implement this efficiently in an ASIC AFAIK, but if there are any hardware designers that disagree please chime in.
503.
by ipsum2 3y ago
There's plenty of use cases, just because you can't think of them doesn't mean there aren't any. Neural network based audio codecs like TSAC, Descript, Encodec, Soundstream are used for music and speech generation, audio
504.
by ipsum2 3y ago
Nothing wrong with critiques.
505.
by ipsum2 3y ago
Edit: 200x more chips, not 20x.
506.
by ipsum2 3y ago
It's not really a lie though. They require 20x more chips (storing all the weights in sram instead of hbm is expensive!) than Nvidia GPUs for a ~2x speed increase. Overall the power cost is more expensive for groq than GPUs.
507.
by ipsum2 3y ago
I no longer work at Meta, but it's most likely 1) the website got hacked and serves malware or 2) false positive. There's plenty of websites that are critical of Facebook. No one at Meta has 4 billion users - no employee is concer
508.
by ipsum2 3y ago
Where do you live where these things don't happen?
509.
by ipsum2 3y ago
I'm skeptical, expect data contamination was the reason for high benchmark scores.
510.
by ipsum2 3y ago
This is awesome! I don't have enough background knowledge to know how to make one myself, do you ever plan to sell these, or make a tutorial on how to build one?

More ›