Gemma-4-26B-A4B does not require 50+ Gb of vram. It is a MoE model so only 4B of active parameters at a time and not as GPU dependent. I can run it on 16gb of vram and ~20gb of DDR5 regular ram for a 8 bit quant.
55 karma · joined December 2, 2016
Looks like the JS is not obfuscated. Even dev comments are still there, which is v cool. Good idea to download it all before they wise up!
wget --mirror --convert-links --adjust-extension --page-requisites --no-parent -e robots=off -P ./website "https://lumon-industries.com/"
wget -P ./website/images "https://lumon-industries.com/images/lumon.png"
wget -P ./website/images "https://lumon-industries.com/images/nope.png"
wget -P ./website/images "https://lumon-industries.com/images/100.png"
wget -P ./website/images "https://lumon-industries.com/images/clipboard.png"
wget -P ./website/images "https://lumon-industries.com/images/mde.gif"
wget -P ./website/shaders "https://lumon-industries.com/shaders/crt.vert.glsl"
wget -P ./website/shaders "https://lumon-industries.com/shaders/crt.frag.glsl"
wget -P ./website "https://lumon-industries.com/favicon.ico"
<script>
navigator.serviceWorker.register?.('/service-worker.js').catch(() => {})
</script>
Finally, download p5 and replace the cloudflare CDN pointing to your own p5.js installation.
Run it all with
python3 -m http.serverAs long as it's niche enough, write about it, and they will come.
Disclaimer, I get XX traffic a week. Nothing noteworthy but it makes me so happy that people read my posts.
I used to regularly report Runescape phishing sites, reporting to the registrar is extremely effective and quite fast
- Github CI pipeline for running tests on a pet project. To the best of my knowledge it was 100% successful with no edits needed, but it was stupid simple.
- Starting place for Jest tests on the same pet project. It didn't give me 100% correct suggestions, but it greatly reduced the mental barrier of using a new technology and writing tests with mocks in an unfamiliar language (typescript)
- Determining how hot a mosfet will get under a certain voltage/amperage with no electrical understanding. I asked for many different mosfets and it got the data sheet numbers frequently incorrect. Super helpful going from 0 to figuring how what to look for on a data sheet and plugging those numbers into to formulas to determine how hot different mosfets would get. (context, this was for a 3d printer)
- Writing a Fresh Prince of Bel Air opening rap scene parody for a board game. I would give it the ideas, it would write the lyrics. Eventually I strung together a bunch of it's lyrics and asked it to make it better. It did by making stuff rhyme better. I had to shorten and change some lyrics to fit to the beat. Way easier than writing it from scratch.
The only downside is that @haugene has gone missing for sometime now, as such there are no firm releases. If you want new features/bug fixes, you have to pull dev, which isn't very ideal. Otherwise the maintainers are doing a great job (shout out to @pkishino!)
I assume your worries come from the leetcode trend in interviews.
Personally, I let recruiters that are knocking down my door via LinkedIn or email my expectations up front. Salary & no leetcode (algorithmic problems), I have found a lot of success. There are many companies willing to give sane interviews with real world SWE problems.
Experience is highly valued in this field. Someone with 10 YOE isn't really competing with someone with 5 YOE. If you talk to anyone involved in hiring nowadays it's extremely difficult to find experienced devs. Experience will always trump education.
- Pre-rendered static files so you can view the site without JS (Front page "terminal" animation and possibly the contact page won't work)
- Blog backed by JSON/Markdown
- Built with Vue
- Hosted on GH pages/backed by Cloudflare
Source: https://github.com/kushagharahi/kushagharahi.github.io
Valetudo is a full-er solution it seems. Props to the author. Though, bumper does not require reprogramming the bot, it does require lots of local DNS hijacking.
The other side of the problem is that lawmakers have no incentive to set aside money/time/resources for pandemic preparedness.
A quote from HHS Secretary under Bush, Mark Leavitt sums up part of the problem nicely:
> “In advance of a pandemic, anything you say sounds alarmist. After a pandemic starts, everything you’ve done is inadequate.”
Another example of lawmakers having no incentive to think about the future:
>The Public Health Emergency Preparedness program’s funding has gone from about a billion dollars in 2003 to $675 million this year, while the Hospital Preparedness program has gone from more than $500 million at its peak in 2004 to less than $300 million today.
> Other large influenza pandemics
> The Spanish flu pandemic was the largest, but not the only large recent influenza pandemic. Two decades before the Spanish flu the Russian flu pandemic (1889-1894) is believed to have killed 1 million people.
> Estimates for the death toll of the “Asian Flu” (1957-1958) vary between 1.5 and 4 million. Gatherer (2009) published the estimate of 1.5 million, while Michaelis et al. (2009) published an estimate of 2–4 million.
> According to a WHO publication the “Hong Kong Flu” (1968-1969) killed between 1 and 4 million people.
> Michaelis et al. (2009) published a lower estimate of 1–2 million.
> The Russian Flu pandemic of 1977-78 was caused by the same H1N1 virus that caused the Spanish flu. According to Michaelis et al. (2009) around 700,000 died worldwide.
> What becomes clear from this overview are two things: influenza pandemics are not rare[...]
Heck, ebola was a decade ago with 300k deaths. Novel viruses appear all the time.
[1] https://ourworldindata.org/spanish-flu-largest-influenza-pan...
Open source, hosted on GitHub pages
You can't search traceroute. Weird.
I find it extremely fulfilling teaching programming and controls to students. It's helped me grow professionally too.
>If a story has had significant attention in the last year or so, we kill reposts as duplicates. If not, a small number of reposts is ok.
That post is from over a year ago, and didn't get much traction.
Blog, Portfolio, Resume, Contact
$6M exit, our shares were bought with the acquiring company's publicly traded shares. Those shares had a 6 month cliff before we could sell them.
~$9 a share at the time of acquisition, ~$4 a share at the end of the 6 months. Sold about 1 year after acquisition for ~$10/share, now at ~$17 a share and climbing.
Made $11k pretax in the end. (+$4k cash gift from founder that got a bigger payout, due to the board not approving more equity for employees pre-acquisition)