xodoh74984 @ xodoh74984 @lemmy.world

Posts

3

Comments

140

Joined

2 yr. ago

It's a 4090 using cublas. I just run the stock llama.cpp server with CUDA support. Do you know if there'd be any advantage to building it from source or using something else?

22466MiB / 24564MiB, awesome, thank you!

I'm close to the limit at 23886MiB / 24564MiB of VRAM used when the server is running. I like to have a bit of headroom for other tasks.

But I'm by no means a llama.cpp expert. If you have any tips for better performance I'd love to hear them!

Example?

Isn't every app that's not open source assumed to be spyware nowadays?

Wake me up when this sort of thing is actually illegal. Preferably punishable with jail time.

Actually, this is a great idea in principle, because the inverse is possible.

We could use Amazon's resources for search, then use a browser plug-in to replace the buy button and have it buy the product from somewhere else using an AI agent.

In the decades that people have been screaming about nuclear plants "taking too long to build" stalling progress, we could have built many nuclear plants and significantly reduced emissions.

We've kicked the can down the road for far too long. We are well past the point of preventing the devastating impacts of climate change we were warned about 20+ years ago.

If this institute is so convinced wind and solar are the answer, I hope they also have vast amounts of lithium pre-mined or a novel approach to energy storage that isn't so damaging.

Reddit natively supports RSS feeds as well. The major feature here is comment support IMO.

Is there any way to fling YouTube videos to SmartTube from a smartphone?

That's the one thing locking me into Kodi.

Training: Creating the model
Inference: Using the model

I feel like people here have forgotten the difference between "vulnerable" and "compromised".

It matters because calling everyone's default setup chat apps compromised implies that an attack has occurred.

Wouldn't it be a vulnerable device? Up until the point it's compromised by downloading a malicious image?

We know

unprecedented

Can't wait for them to get SLAMMED by Putin for something something click bait

One of the best use cases for LLM's is to desensationalize the news. I'm tired, Mom.

That sounds much more polished than the earlier episodes. Glad I was able to help!

Awesome! Happy to help!

I'm referring to streaming all audio from the phone over WiFi in the same way that Bluetooth audio can be streamed.

Cast is occasionally supported by specific apps, but there is no audio driver level support for any protocol.

DLNA / UPnP streaming is the most open and widely supported WiFi based audio protocol I'm aware of. Many modern stereo receivers can receive DLNA audio streams. Bluetooth has always had limitations both with range and quality.

For the user whose VRAM knob goes to 11

Here are some techniques for measuring alignment:

https://arxiv.org/pdf/2407.16216

By in large, the goals driving LLM alignment are to answer things correctly and in a way that won't ruffle too many feathers. Any goal driven by human feedback can introduce bias, sure. But as with most of the world, the primary goal of companies developing LLMs is to make money. Alignment targets accuracy and minimal bias, because that's what the market values. Inaccuracy and biased models aren't good for business.

I believe it's supported. On initial install, I remember reading that it was possible to fling a video to SmartTube from the official YouTube app, but that wasn't an option for me using NewPipe on Android. Saying that out loud, it's probably worth it for me download the official YouTube app, just so I can do that.

Also, the GitHub readme is a bit long and intimidating for installation. All you have to do is install this app:
https://www.aftvnews.com/downloader/

Then use the code 28544 to install SmartTube.

It's just a way to sidestep Google, since Google obviously doesn't want an app like this listed on their official store.