Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Facebook LLAMA is being openly distributed via torrents

github.com

Post reply on HN
Print view View original

281–290 of 719 posts

Re: Facebook LLAMA is being openly distributed via torrents

#281

Post by dx034 »

renonce wrote: post #180
ot wrote: post #16

In case it's not clear what's happening here (and from the comments it doesn't seem like it is), someone (not Meta) leaked the models and had the brilliant idea of advertising the magnet link through a GitHub pull request. The part about saving bandwidth is a joke. Meta employees may have not noticed or are still figuring out how to react, so the PR is still up. (Disclaimer: I work at Meta, but have no relationship w…

It's fairly easy to obtain the weights. I've had two of my friends downloading these weights and sharing them with me, so it's probably not surprising that the weights got leaked.

Was every Meta employee able to download them or why did so many people have access?

Re: Facebook LLAMA is being openly distributed via torrents

#282

Post by mavhc »

Retr0id wrote: post #97
sebzim4500 wrote: post #92

Earlier quoted context omitted.

I'm in the UK and can view that link without a VPN.

Consider yourself very lucky that your ISP doesn't suck. I'm in the UK and the link won't load, due to TLS SNI filtering (Virgin Media).

TOR browser fixes all

Re: Facebook LLAMA is being openly distributed via torrents

#283

Post by Yoric »

oefrha wrote: post #160
moffkalast wrote: post #109

Earlier quoted context omitted.

"They turned the model into a pickle? Funniest shit I've ever seen." But seriously, why not something more human readable and text-based if it's just weights?

Because human-readable text-based formats are really inefficient to both download and load, especially when in the hundreds of GB range. And no human cares to read billions of weights.

Agreed. However, there are much better formats than Python pickles for exchanging binary data. As it is, using PyTorch means that you force your users to also use PyTorch, which is a shame, as libtorch (which is what makes PyTorch work) offers a much more portable format (which I suspect might also be more efficient at least in terms of raw size, but I haven't checked).

Re: Facebook LLAMA is being openly distributed via torrents

#284

Post by ironrabbit »

Stagnant wrote: post #277
narrator wrote: post #151

Earlier quoted context omitted.

It's funny that part of the 4chan excitement over this is that they think they'll get back the AI girlfriend experience of when character.ai was hooked up to uncensored GPT-3. All that has been thoroughly shut down by character.ai and Replika and they just want their girlfriends back.

Correct me if I'm wrong but doesn't character.ai use their own model and isn't associated with OpenAI? At least I can't find any information that would claim so. Anecdotally, as a roleplaying chat experience, char.ai seems to perform way better than anything else publicly available (doesn't get repetitive, very long memory). It also feels different to GPT3 on how it is affected by prompts. I've just assumed that char…

Character has their own models, and anecdotally I've heard they have one of the better LM training codebases out there.

Re: Facebook LLAMA is being openly distributed via torrents

#285

Post by __MatrixMan__ »

sofixa wrote: post #240
bentcorner wrote: post #224

Earlier quoted context omitted.

I posit to a friend that: a) As these AI constructs become more advanced (especially around memory and personalization), we will eventually be able to treat them as people b) Some business will eventually sell an off-the-shelf product (hardware and/or software) that is an AI you can bring into your home, that you can treat as a friend, confidant and partner c) Someone will eventually lose their AI friend of many mont…

> a) As these AI constructs become more advanced (especially around memory and personalization), we will eventually be able to treat them as people There already planned products to "capture" someone's voice and personality to be able to continue experiencing "them" after their death? Shit is already weird. https://technode.global/2022/10/21/this-startup-allows-you-t...

> One of its products, Re;memory, is a virtual human service based on AI technology which recreates the clients’ late family members by recreating their persona – from their physique to their voice. The service is for those that wish to immortalize their loved one’s story of life through a virtual human.

There's so much sci-fi about this, it's pretty well charted territory. I bet reality will find a twist at haven't thought of though.

Re: Facebook LLAMA is being openly distributed via torrents

#286

Post by ImprobableTruth »

vintermann wrote: post #175
eigenvalue wrote: post #60

I'm not surprised-- I recently suggested that someone might try to pull an Aaron Swartz with the LLAMA weights (i.e., release them in an uncontrolled way similar to how Aaron attempted to release the JSTOR database). It's quite misleading for FB to claim that they are being so open, but then hoard the weights and only release it to a few academics. If the paper is to be believed, this is a major development, allowing…

It was already the most open language model in its class, given that the code for training and inference was available and it only used public data for training. For Google and OpenAIs offerings, have fun reimplementing it from descriptions in the paper (including small crucial details that they may have left out), training it for a month, and then wondering if the implementation or the training data is the reason yo…

The training code is not available.

Re: Facebook LLAMA is being openly distributed via torrents

#287

Post by seydor »

candiddevmike wrote: post #104
seydor wrote: post #45

Earlier quoted context omitted.

> massively expensive to produce and self-contained to a degree that you can distribute the end product in a torrent. So, like movies or software

Or a car

Where s the download link?

Re: Facebook LLAMA is being openly distributed via torrents

#288

Post by ProjectArcturis »

rnk wrote: post #265
dmix wrote: post #230

Earlier quoted context omitted.

Reminds me of Ray Kurizwell's obsession with uploading brains to the cloud to get back his beloved father.

When we can upload our brains to the cloud, and you can do something with them like interacting or running the brain, then we'll all be effectively immortal. That's a pretty big deal. See the book altered carbon.

They wouldn't be us. We will still die when our bodies fail. But maybe there will be some AI tricking our friends and family into thinking we're still there.

Re: Facebook LLAMA is being openly distributed via torrents

#289

Post by HarHarVeryFunny »

WithinReason wrote: post #52
rnosov wrote: post #2

Just to make it clear, does this torrent include model weights?

Folder structure for the 2 smaller models look like this: LLAMA │ tokenizer.model │ tokenizer_checklist.chk │ ├───13B │ checklist.chk │ consolidated.00.pth │ consolidated.01.pth │ params.json │ └───7B checklist.chk consolidated.00.pth params.json

So what is content of those various files? Does this include the full models themselves, or just the weights ?

Re: Facebook LLAMA is being openly distributed via torrents

#290

Post by renonce »

dx034 wrote: post #281
renonce wrote: post #180

Earlier quoted context omitted.

It's fairly easy to obtain the weights. I've had two of my friends downloading these weights and sharing them with me, so it's probably not surprising that the weights got leaked.

Was every Meta employee able to download them or why did so many people have access?

Anyone can submit a request form and enter their email address to request access to the weights. People who have a .edu email and are involved in deep learning are likely to succeed and get a download link sent to their email.
Post reply on HN