Some phones have limiters to keep the battery at 60%-80%.

I believe most can do this with the right software.

It's not a fix, but it should extend the life considerably.

Another benefit: modern smartphones have large GPUs, large media blocks, and fast RAM.

With the right software, they can be a surprisingly powerful AI host or transcoding server.

Real world GPU performance is hugely influenced by hand optimization of the CUDA kernels.

Power/Weight is extremely high. A tiny wankel will do the job, and weight is everything on cars.

It does prefer a narrow RPM band, which is fine.

Reliability is the biggest concern TBH, but maybe that's not a huge bummer if its more of a backup/assistant engine.

Reliability for rotaries hasn't been a concern for a long time. Modern apex seals work well and last a reasonably long time. There is a need to stop parroting facts from the 1980s.

I would note the actual leading models right now (IMO) are:

- Miqu 70B (General Chat)

- Deepseed 33B (Coding)

- Yi 34B (for chat over 32K context)

And of course, there are finetunes of all these.

And there are some others in the 34B-70B range I have not tried (and some I have tried, like Qwen, which I was not impressed with).

Point being that Llama 70B, Mixtral and Grok as seen in the charts are not what I would call SOTA (though mixtral is excellent for the batch size 1 speed)

Miqu is a leaked model -- no license is provided to use it. Yi 34B doesn't allow commercial use. Deepseed 33B isn't much good at stuff outside of coding.

So it's fair to say that DBRX is the leading general purpose model that can be used commercially.

Model weights are just constants in a mathematical equation, they aren’t copyrightable. It’s questionable whether licenses to use them only for certain purposes are even enforceable. No human wrote the weights so they aren’t a work of art/authorship by a human. Just don’t use their services, use the weights at home on your machines so you don’t bypass some TOS.

Photographs aren't human-made either, yet they are copyrightable. I agree that both the letter and the spirit of copyright law are in favor of models not being copyrightable, but it will take years until there's a court ruling either way. Until then proceed with caution and expect rights holders to pretend like copyright does apply. Not that they will come after your home setup either way.

Photos involve creativity. Photos that don't involve creativity aren't usually considered copyrightable by the courts (hence why all copyright cases I followed include arguments that establish why creativity was a material to creating the work).

Weights, on the other hand, are a product of a purely mechanical process. Sure, the process itself required creativity as did the composition of data, but the creation of the weights themselves do not.

Model weights are effectively public domain data according to the criteria outlined in statement issued by the US copyright office a year ago: https://www.federalregister.gov/documents/2023/03/16/2023-05...

Qwen1.5-72B-Chat is dominant in the Chatbot Arena leaderboard, though. (Miqu isn't on there due to being bootleg, but Qwen outranks Mistral Medium.)

Yeah I know, hence its odd I found it kind of dumb for personal use. Moreso with the smaller models, which lost an objective benchmark I have to some Mistral finetunes.

And I don't think I was using it wrong. I know, for instance, the Chinese language models are funny about sampling since I run Yi all the time.

It's Deepseek, not Deepseed, just so people can actually find the model.

For all the Model Cards and License notices, I find it interesting there is not much information on the contents of the dataset used for training. Specifically, if it contains data subject to Copyright restrictions. Or did I miss that?

Yeah, its an unspoken but rampant thing in the llm community. Basically no one respects licenses for training data.

I'd say the majority of instruct tunes, for instance, use OpenAI output (which is against their TOS).

But its all just research! So who cares! Or at least, that seems to be the mood.

The conspiracy theorist in me says thats a low priority due to perverse incentives (namely selling more storage at a huge markup).

Another rationale is that the Apple ecosystems tends to not use JPEG by default anyway, right? It uses HEIC or something.

Yes but most of images are from Whatsapp messages. And they are JPEG.

Apple was always very frugal regarding memory and storage

Being a "hero" open source dev for a project like that can require a lot of neuroticism.

Sometimes it works, but sometimes the project is just too big, I think.

It's not either or, you can use different vendors for different tasks.

tinygrad isn't in the realm of production ready though, AFAIK.

Yes but the same could be said about rocm.

The MI300 is the best accelerator you can buy, for many current workloads.

It's technically way more advanced. Not as outrageously priced as an H100 either.

I think you are preaching to the choir, and AMD is not listening.

AMD would be selling 48GB 7900s or AI-only W7900s if they really wanted a consumer card ramp.

They don't. Not because they can't (they literally prevent OEMs from doing so, who would double up VRAM in a heartbeat without AMD lifting a finger), but because AMD doesn't want that.