181–190 of 335 posts
I am curious if it can generate audio for a country. For example, the sirens sound in the sample doesn't like I siren I would recognise. Sounds like an American one?
> MusicGen, which was trained with Meta-owned and specifically licensed music, generates music from text-based user inputs, while AudioGen, which was trained on public sound effects, generates audio from text-based user inputs. Meta is really clearly trying to differentiate themselves from OpenAI here. Open source + driving home "we don't use data we haven't paid for / don't own".
I'm looking forward to playing with the M1 Mac apps/cli-tools that will probably come out for this in the next week or so! Being able to run this stuff locally is a lot of fun.
Are the M1 macs capable enough? I'm eyeing and upgrade in the coming months and I'm curious if a MacBook would be suitable
Earlier quoted context omitted.
In copyright law the use of the work itself is considered a commercial benefit, so "noncommercial use" is an oxymoron. Consider these situations: - If I use AudioCraft to post freely-downloadable tracks on my SoundCloud, I still get the benefit of having a large audio catalog in my name, even if I'm not selling the individual tracks. I could later compose tracks on my own and ride off the exposure I got from posting…
This is a weird comment. Do you think that non commercial use simply doesn't exist or something? Because non commercial use isn't some crazy concept. It is a well established one, that doesnt disclude literally everything. Also, you are ignoring the idea that Facebook will almost certainly not sue anyone for using this for any reason, except possibly Google or Apple. So if you aren't literally one of those companies…
Anyone know if there are ways, as-is, to speed this up on Apple Silicon? This setup takes 5 minutes: - Mac Studio M1 Max 64GB memory - running musicgen_app.py - model: facebook/musicgen-medium - duration: 10s
>.if torch.cuda.device_count():
>. device = 'cuda'
>. else:
>. device = 'cpu'
At least that is what I ran into last time I tried to speed this up on an M1. It's possible there are fixes.
Anyone know if there are ways, as-is, to speed this up on Apple Silicon? This setup takes 5 minutes: - Mac Studio M1 Max 64GB memory - running musicgen_app.py - model: facebook/musicgen-medium - duration: 10s
I see from the musicgen.py- >.if torch.cuda.device_count(): >. device = 'cuda' >. else: >. device = 'cpu' So pytorch will fall back to CPU on a Apple Silicon. Ideally it would use Metal for acceleration (MPS) instead of just plain 'CPU', but if you replace CPU with MPS you'll probably run into a few bugs due to various Autocast errors and I think some other incompatibility with Pytorch 2.0. At least that is what I ra…
I’ll have to check again, but I remember AFAICT my hardware wasn’t getting saturated, so maybe there’s headroom for mac cpu performance. And of course in the meantime I’ll be refreshing the ggml github every day
Earlier quoted context omitted.
In copyright law the use of the work itself is considered a commercial benefit, so "noncommercial use" is an oxymoron. Consider these situations: - If I use AudioCraft to post freely-downloadable tracks on my SoundCloud, I still get the benefit of having a large audio catalog in my name, even if I'm not selling the individual tracks. I could later compose tracks on my own and ride off the exposure I got from posting…
This is a weird comment. Do you think that non commercial use simply doesn't exist or something? Because non commercial use isn't some crazy concept. It is a well established one, that doesnt disclude literally everything. Also, you are ignoring the idea that Facebook will almost certainly not sue anyone for using this for any reason, except possibly Google or Apple. So if you aren't literally one of those companies…
As for "Facebook won't sue"? Sure, except we don't have to worry about just Facebook. We have to worry about anyone with a derivative model. There's an entire industry of copyleft trolls[0] that could construct copyright traps with them.
Individuals can practically ignore NC mainly because individuals can practically ignore most copyright enforcement. This is for the same reason why you can drive 55 in a 30mph zone and not get a citation. It's not that speeding is now suddenly legal, it's that nobody wants to enforce speed limits - but you can still get nailed. The moment you have to worry about NC, there is no practical way for you to fit within its limits.
[0] https://www.techdirt.com/2021/12/20/beware-copyleft-trolls/
For my two cents; the goal of our human pursuit is to advance our technologies and systems to the point we can all live carefree lives where the focus of our pursuits is self defined. It is not to "own more rights" to shit. Sorry, but no one actually owns what they create, that's the point of creation. And if you take issue with the term create, that only reinforces my point. We're all influence machines, input and o…
Agreed!
> the future should not be about preserving some peoples rights to limit our collective advancements over their personal wants.
Systems that don't allow people to extract value from the hard work they put into collective advancement do not seem to lead to collective advancement over time and at scale. Incentives matter. No one's going to spend all day making candy and put it in the "free candy" bowl when that one asshole kid down the street just takes all of the candy out of the bowl every single day.
At small scales (i.e. relatively few participants with a relatively high number of interactions between them) then informal systems of reciprocity and reputation are sufficient to disincentivize bad actors.
At large scales where many interactions are one-off or anonymous, you need other incentives for good-faith participation. There's a reason you don't need a bouncer when you have a few friends over for drinks, but you do if you open a bar.
For my two cents; the goal of our human pursuit is to advance our technologies and systems to the point we can all live carefree lives where the focus of our pursuits is self defined. It is not to "own more rights" to shit. Sorry, but no one actually owns what they create, that's the point of creation. And if you take issue with the term create, that only reinforces my point. We're all influence machines, input and o…
If nobody is beholden to any job or duty, and the machines do everything, who is to say I don't want to make every machine on earth dance in a flash mob? I cannot do that, because it would require other people to halt their use of the machines. Abundance is a false promise and one we should be quick to shoot down lest we surrender our future rights to the ones advertising it.
Removing the worth of people in their jobs removes their leverage in the constant resource allocation negotiation in the economy. Given that we just witnessed Elon Musk spend 20,000 average American lifetime earnings worth of wages just to be the new dictator of a social media company, I'm not sure that I want those negotiations to take place only among the giga-rich.
Earlier quoted context omitted.
My understanding (IANAL) [1] is that copyright licenses have no say on the output of software. Further, CC licenses don't say anything about running or using software (or model weights). It's therefore questionable whether the CC-BY-NC license actually prevents commercial use of the model. [1] https://opensource.stackexchange.com/questions/12070/allowed...
You're correct, but no one has had the balls (or the lawyers) to clarify this in court yet. Expect to see hosting providers complying with takedown requests for the foreseeable future.