thewataccount

13 ms·

31.
by thewataccount 3y ago
I feel like I saw one once but lost it - Is there a githubrepo/tutorial for how linear algebra is used for a very small model just to demonstrate how that allows it to "learn"? I've got the calc, I just don't unders
32.
by thewataccount 3y ago
Can anyone suggest a something that teaches Linear Algebra with a practical applications, especially for software engineering? I can sorta kinda get the theory, but every demonstration involves moving an arrow around which is.... not someth
33.
by thewataccount 3y ago
That's for the non-finetuned GPT3.5turbo model. Finetuned is 1.2cents/1k in and 1.6cents/1k out. So it'll likely be closer to 2cents depending on what you're doing. I'm not saying it's not useful, at 2c pe
34.
by thewataccount 3y ago
Totally understand! I'm sure you'd get swarmed with requests. Are you interested in non-LLMs as well? Stable diffusion for example?
35.
by thewataccount 3y ago
Hello, I think this is a great project! If an individual is looking to contribute to the field with different training data ideas - would they need to first establish themselves and get your attention, or would there be a way to submit a pr
36.
by thewataccount 3y ago
> fool kernel level anti-cheat That's the problem though - if the compatibility layer can fool the anti-cheat then so can actual cheats. Both cheats and compatibility layers do the same thing - modify behaviour. Generally anticheat
37.
by thewataccount 3y ago
I'll argue that python is very alive in the system management/scripting space. Because of the library support (especially for data) it's also popular as api servers. It's not as popular for entire applications but is sti
38.
by thewataccount 3y ago
> My understanding is there is no technical reason these anti-cheats cannot work via Proton, but the anti-cheat makers will ban if they figure out you're not on Windows out of some weird fear of people having root access to the syst
39.
by thewataccount 3y ago
I'm not 100% sure they're talking about this specifically but logit control/manipulation is often used to conform to a specific schema. https://github.com/guidance-ai/guidance I'm going to butcher t
40.
by thewataccount 3y ago
I think what you're trying to say is that just because someone - especially [large company] - tries to improve/integrate popular open projects doesn't mean it's always EEE. Which I doubt EEE is purposeful the majority of
41.
by thewataccount 3y ago
I think they're suggesting simply doing +2 +1 is not a good idea since ~half of all numbers are effectively non-prime simply by being even numbers. You can double the speed by using +2 without using any fancy tricks, just changing a si
42.
by thewataccount 3y ago
Refact has worked for me. Hopefully exllama will support CodeLlama.
43.
by thewataccount 3y ago
Any plans to support IntelliJ?
44.
by thewataccount 3y ago
Is automation not what every engineer strives for when possible? Especially software developers. From my experience with github copilot and GPT4 - developers are NOT going anywhere anytime soon. You'll certainly be faster though.
45.
by thewataccount 3y ago
Unfortunately what I tested was StarCoder 4bit. We really need exllama which should make even 30b viable from what I can tell. Because codellama is llama based it may just work possibly?
46.
by thewataccount 3y ago
For in-editor like copilot you can try this locally - https://github.com/smallcloudai/refact This works well for me except the 15B+ don't run fast enough on a 4090 - hopefully exllama supports non-llama models, or
47.
by thewataccount 3y ago
> Let's assign blame to Nvidia, though. I'm not going to excuse Nvidia - their driver is buggy and they're still missing DLSS3. Many of the problems I had were nvidia specific, especially anything touching wayland. However
48.
by thewataccount 3y ago
281M and 235M param models too. https://github.com/facebookresearch/seamless_communication/b... I don't really know how the metrics they list compare to whisper, I'm very curious if these are fast enough
49.
by thewataccount 3y ago
I've used linux as my desktop for 2+ years. The games I play like CSGO work very well*, you have a steamdeck so you know how game support especially with anti cheat is. However, I've had many bugs. Nvidia on linux is painful. Just
50.
by thewataccount 3y ago
Long running processes only need to load their imports, configuration, database connections, etc. once, and don't incur overhead for each request. On top of this, things like imports get to be cached, and there's less need for a d
51.
by thewataccount 3y ago
You're looking for "HumanEval" tests. Not saying this is the best way to test it, but it's the only standard test I know of that code models are compared with and are commonly benchmarked for The current best models you&
52.
by thewataccount 3y ago
I have some bad news for you about Tylenol/Paracetamol.... The science is "We did many double blind studies on a large population and observed these effects to a degree statistically significant compared to placebo" We can re
53.
by thewataccount 3y ago
> It's really nice when every compiler/language is on the latest version by default. I've seen breaking changes in my dependencies from 3.7 -> 3.8 -> 3.9 -> 3.10, both at work and at home. I've also had cons
54.
by thewataccount 3y ago
Its worth noting that from what I can tell - A model well trained in most languages would be able to learn the niche ones much more easily. The vocab size of llama2 is 32,000. I guess I personally don't think that there's enough d
55.
by thewataccount 3y ago
I checked a little bit ago but I might have missed it so please correct me. Are there plans to expand self-hosting support? The migrations are a big step forward. Are you intending to fill the roll of being a framework akin to a "super
56.
by thewataccount 3y ago
Check out the hugging face leader board and try one of the finetunes. I can't run llama2 70B - but I remember llama1-Guanaco30B was a major leap over the base llama30B model for me.
57.
by thewataccount 3y ago
Any chance of signed builds returning? It's bizarre to me that we would move _away_ from signed builds. 2FA means we can trust the person that logged in - but we still don't trust that PyPI is being honest (no offense).
58.
by thewataccount 3y ago
I think simply having the vocab more code friendly (e.g. codex) would make the biggest difference, whitespace is the biggest one (afaik every space is a token), but consider how many languages continue `for(int i=0;`, `) {\n`, `} else {`, &
59.
by thewataccount 3y ago
I get decent performance with my 4090, enough that LLMs with exllama at 30B quantitized are very usable. But we're severely VRAM limited, especially on lower end hardware which rarely sees > 10GB of VRAM. I don't know how much
60.
by thewataccount 3y ago
I'm not excited about developing "in the cloud". Privacy, speed, lock-in, etc. Especially with google's track record.

More ›