Aissen

12 ms·

181.
by Aissen 2y ago
> Or at least she did "Internet is new territory, uncharted territory to all of us." (the infamous Neuland).
182.
by Aissen 2y ago
> The 'customer support of last resort' genre is common and not usually a good fit for HN They're already off Cloudflare, I would see this story more as "Dealing with tech company X is a business risk" cautionary
183.
by Aissen 2y ago
Infrastructure at scale is not as expensive as cloud vendors would have you think. Reaching scale and being a first-mover gives a significant advantage, because deploying a CDN with caches well connected with(in) every major ISP around the
184.
by Aissen 2y ago
There are open source apps for that you can find on f-droid.
185.
by Aissen 2y ago
Nice to find privacy leaking clients: https://www.caniemail.com/features/css-pseudo-class-visited/
186.
by Aissen 2y ago
Obviously, you discover and talk to AI services with consul, store AI secrets in vault, deploy AI platform nomad with AI infrastructure software terraform.
187.
AltStore Pal: First Alternative iOS App Store for EU(rileytestut.com)
1 points by Aissen 2y ago|1 comments
188.
by Aissen 2y ago
How much memory would inference take on this type of model? What's the impact of it being an MoE architecture?
189.
by Aissen 2y ago
This is great for entertainment (and hopefully the main application), but we need clear marking of such type of videos before hallucinated details are used as "proofs" of any kind by people not knowing how this works. Software vid
190.
by Aissen 2y ago
Thanks, that's very interesting. Could you share which areas of business you plan on exploring?
191.
by Aissen 2y ago
Isn't it a bit of small round for a hardware startup? It's basically the amount of money put in a single 10000 units batch of the latest framework 16. Multiply that by ~3 to account for lower effective BOM, and I still find this l
192.
by Aissen 3y ago
It is true for Sonic 1, but Sonic 2 is infamously difficult on Game Gear, because the levels where designed with the Master System in mind, and it didn't port well to the smaller Game Gear resolution (160x144 vs 256x192): the assets ar
193.
by Aissen 3y ago
> I guess by training time they meant with respect to downtime of the hardware, not utilisation of the hardware potential. That was my understanding as well, those are two different levels of efficiency.
194.
by Aissen 3y ago
I'm curious, do you have any pointers? This is what this article mentions: > Those improvements resulted in an overall effective training time of more than 95% https://ai.meta.com/blog/meta-llama-3/
195.
by Aissen 3y ago
> very long time On one of Meta's 24k H100 clusters running at 95% efficiency, it's 2.3 days.
196.
by Aissen 3y ago
Thanks, I was looking for information on this, it seems to be lower speed than pure-CPU inference on M2, and probably much worse than a ROCm GPU-based solution?
197.
by Aissen 3y ago
A quick search into it shows that this Ryzen AI NPU's support isn't integrated into upstream inference frameworks yet — so right now it's just useless silicon surface you pay for :-/
198.
by Aissen 3y ago
Reason: https://twitter.com/WizardLM_AI/status/1780101465950105775 (they forgot toxicity testing before release and are doing it now)
199.
by Aissen 3y ago
Deleted. Archive: http://web.archive.org/web/20240415221214/https://wizardlm.g... https://github.com/nlpxucan/WizardLM/commit/4218c0ba64a84cfd...
200.
by Aissen 3y ago
Did you try on both Firefox and another chromium-based browser to compare? If it works similarly for you, I might take more time to investigate why it's janky here. Edit: I tested with nightly (so a new profile), and performance is sim
201.
by Aissen 3y ago
Phanpy is extraordinary, and my main client. Unfortunately, it pushes mobile browsers to the limit. The scroll performance in Firefox for Android is abysmal, and deep threads can make Chrome choke on rendering. Despite all that, I keep usin
202.
by Aissen 3y ago
If you're interested in truly open source LLMs, I've been playing CroissantLLM, a bilingual French/English LM; it's smaller than OLMo (1.3b vs 7b, so not in the same category), and has checkpoints, training code, logs, d
203.
by Aissen 3y ago
> blessed Ubuntu version with the blessed kernel version To an SRE, this is a nightmare to read. Cuda is bad in this regard (can often prevent major kernel version updates), but this is worse.
204.
by Aissen 3y ago
Exploit a memory safety issue in the tokenizer/or other parts of your LLM infra written in a native language.
205.
by Aissen 3y ago
And I thought France had it bad… There are a lot of hoops to jump through when owning a company in France, but the creation is the "easy" part. And we got rid of the minimum capital many years ago. A good accountant will even guid
206.
by Aissen 3y ago
It was not the first instance of this happening; other times I'm not certain it was git's fault.
207.
by Aissen 3y ago
It's not a minor note, it's a major reason that the github auto-generated tarballs are useless as-is, since they are not stable.
208.
by Aissen 3y ago
Have you never spent hours (or even days!) debugging an issue that seemed obvious in retrospect, once all incorrect assumptions have been eliminated?
209.
by Aissen 3y ago
Looks like one of the backdoor authors even went and disabled the feature the exploit relied on directly on oss-fuzz to prevent accidental discovery: https://social.treehouse.systems/@Aissen/112180302735030319 https:&
210.
by Aissen 3y ago
It's not what I wanted to read, but thanks for the honest answer! I realize creating a new toolkit is a huge endeavor, and I hope you'll some point reconsider :-)

More ›