HN user

bart__

49 karma
Posts0
Comments36
View on HN
No posts found.

I had similar ideas in the past but never made anything. I wanted to have a host mode where it is hosted on a Chromecast. I believe Chromecast can run custom JS so it should be possible. Maybe even rendering on a phone and streaming to the Chromecast. That would make it even more low friction, so nobody needs to get a tablet or laptop out and hook it up to the screen. Would you be interested in adding that? If you'd want I can maybe even help out with implementing it. Maybe somebody mentioned this before but I couldn't find it.

The best solution I could think of on the spot was this: First write down every license place until it fills up. Then for each time a car drives by, give it a 1/2 chance to write it down. Moving down the list regardless of the outcome. Then once the bottom is reached, do it again but with 1/3 a chance. Over and over again, decreasing the chance each time.

I checked for which situations it is valid, and it surprisingly was when there are many more cars than the list size. But it is not valid when there are just a few more cars than the list size.

4090 only has 24 GB and will only be able to fine tune (and merge, which is more memory intensive) the 7B model. The RTX6000 with 48 GB is able to fine tune the 13B model. The 70B model presumably needs multiple GPUs, like 4 RTX6000. For people starting out, you can also use a free GPU from Google colab to fine tune a 7B model. Finetuning 70B gets more expensive and I would suggest trying smaller models first with a high quality dataset.

It is mostly linear I think.

~20 GB vram for the 7B model and 48 GB for the 13B model. It depends on the context size as well. I'd recommend renting a 4090 from a cloud provider like runpod/vast ai to get started, using a PEFT tutorial.

LibrePCB 0.1.7 4 years ago

Totally contra to this project, but relevant for your point. I found EasyEDA, developed by the people from JLCPCB, to be very easy to use for hobbiest boards. Their integration with their SMT assembly service was also pretty neat for my usecase

On your second point: The paper shows, when comparing recovered to recovered with single dose, that there are about half as many re-infections. And yes we're talking about 0.3% vs 0.15% reinfection (that's why there is not enough power for significance), but halving the chances may still be important. I'm opposed to compulsory vaccination either way though.

I did not know that, honestly quite surprised! One thing to keep in consideration is that there is a very big evolutionary pressure for covid to get a mutation that bypasses the vaccine antibodies and still infect others, given that there are so many active infections still out there. I now have good hope though!