Torrents should really be the preferred method for distributing AI model weights. Why rely on a single point of failure like Hugging Face? BitTorrent was made for exactly this.
In my experience public torrents often die as they grow older. It doesn't help that BitTorrent V1 makes long term seeding annoying, and BitTorrent V2 is almost never used.
I never understood this, is there anything that makes it difficult for the original uploader, the one that supposedly offers the file directly, to offer a torrent instead for the same amount of time?
As far as perennity is concerned it seems strictly better.
Distros are a bad use case for P2P anyway since you depend on upstream as soon as you start upgrading and installing packages.
This is true for any distribution method not just p2p. You can even download a nightly through torrents so what does it matter how the data is transferred if it’s always going to require `apt update`?
Yeah, I just use the "netinstaller" ISOs since it's much smaller and never needs to be updated. If I had a need for air-gapped/offline installs I'd either download a larger ISO or just manually install packages from .deb as needed.
You can trivially have storage deduplication for the files served via torrent, transparent to the protocol. The most trivial version of this that you can do today with pretty much any client is having a single directory containing files serving multiple overlapping torrents.
I don't see why it would be. It's transparent to other clients just like it is to the protocol. It cannot be more complex than alternatives by construction.
Yes, models are a good use for P2P especially if everyone agrees to share the same torrent and someone (or a cohort) commit to seeding for the long haul.
The problems begin with taking 5-10 minutes to locate a file on the network. That's right, when you ask for a file it takes 5-10 minutes. Also if the file isn't in the network at all then it never terminates.
Nobody noticed because everyone just used the central web gateway that cached every file anyone ever accessed.
Because history is path-dependent, as engineers keep learning over and over again. It doesn't matter whether Plan9 is theoretically superior to Linux - we're all on Linux and nobody's porting all the apps over.
The biggest problem with BTv1 was the lack of per-file checksumming, and swarm merging (i.e. individual files have shared seeding pools across torrents). BTv2 specs the latter, but I think only BiglyBT actually implements it. Having both of those features from the get-go would've gone a LONG way to fixing the dead torrent problem.
ages ago I tried using IPFS to more or less accomplish this, I imagined it to act more like a weights/training data network fs that everyone would be able to participate in.
HF is meant to be a single point of control. AI models and Linux distros aren't usually for normies, so distribution via torrents would make sense, especially to save the provider some bandwidth. Ubuntu has been offering torrent downloads for ages. No mention of torrents on HF. I believe most downloads will soon be account/EULA-walled.
great initiative, it's really weird seeing efficiencies get rediscovered in the LLM audience, because these efficiencies aren't even what I would consider to be old
Honestly, models are torrents will end any effort from the big AI labs to stop open models. No way to prevent weights from being shared, just like mobies. Genie is out of the bottle
Same here. Inspired by both, I used to delivery videos over torrent to in door machines since at that time, there's no CDNs (or it was difficult to get one).
I've seen this used for distributing container images in networks with awkward network topologies (e.g. a lot of bandwidth within a site or sub-site but limited bandwidth to central registries)
Once I was using Blizzard's downloader to install something (StarCraft, Diablo, I don't remember), and it was kinda slow. I disabled P2P downloads and speed skyrocketed, and I said "Huh, this was unexpected".
When P2P downloads disabled you could see the list of CDNs you're downloading from and mine had a single IP on that list. It looked familiar. Then it dawned on to me. It was the Akamai server which we were hosting in our system room, at 15 minutes of driving distance. After a chuckle, I went to get a cup of tea, because that was entertaining than the game itself.
Then of course, I dived into whatever I was installing that night.
Edit: From the screenshots in the wiki, I remembered that the progress bar was red. It was possibly Diablo 3, then. However, I'm still not 100% sure about it.
So how come was using your own ip? Become a Diablo node installer so you downloaded from there, like how can they convert your own akamai instance into anode without you knowing?
It's not my own IP, but our IP block. As an institution we have a couple of big networking related hats, so we have a couple of B blocks (/16 networks) for ourselves. The IP was from one of these blocks.
However, these network operations (and CDN related stuff) are not managed by us, but by a couple of high profile guys next door. I knew we had an Akamai node, because the machines were shipped recently and we had a chat about it.
Funnily, some of the performance optimizations on Akamai's disk access algorithms (SSDs were not dime a dozen back then, so people were still optimizing for spinning drives) was apparently developed by a friend of mine, but I don't know any details about that.
Several companies tried this for distributing software.
It was very controversial. Users were angry that software companies were using their internet bandwidth to distribute their software. Made a lot of people angry.
As someone who was annoyed by it at the time, I don't think it's a good example of that.
This was in an era of much more limited upstream bandwidth. The p2p nature was obscured from non-technical users in some implementations. So on and so on. I'm sure there are plenty of writeups from that era detailing why it was a bad idea.
These companies already used capable CDNs to distribute patches. It was a way for them to shift their CDN bill onto their paying customers.
When you have barely the upstream bandwidth to send the ACK stream of your downloads, anything sending something becomes a big burden on your system.
Not everyone has the luxury of fully symmetric internet connections even in this age. Yes, my uplink speeds are absurd when compared to a decade ago, but when I ratio it to my downlink, it's still slow.
I have no comment on residential proxies. Somebody doing something in my name is unacceptable to put it very mildly.
I’ve been wondering when this will come. The days are numbered for abliterated models to be published on HF I think. Why wouldn’t the government want a central control there?
If I search for 'uncensored' there are no torrents available. Uncensored models should be top priority, especially now that Nvidia owns HuggingFace and will enshittify the platform in accordance with upcoming US laws.
There are such results on huggingface tho. Also search for the keywords “abliteration” and “heretic”. Heretic is a tool used to abliterate, that is decensor, models
That was my my first search too. I know they are still listed on HF but every time I try and use one, the links are dead or the size is beyond my scope. Was hoping for a fresh batch. Ill check back.
Torrents always seemed like the more sensible way to distribute model weights.
Though I have been disappointed that most of these have been spurred on by the misleading claim that abliterated models were being taken down from HuggingFace because they removed an abliterated model. HF took one abliterated model down because the uploader was spamming people who requested access with sketchy requirements to pay for it.
Plus, there are a bunch of these types of sites, all of them have a couple of models and otherwise completely dead.
There's also the problem of catching malicious models that have been fine tuned to exfiltrate credentials. It would be nice to have means of checking hashes against the HF versions (or against other reputable sources). I'm guessing this is probably easy when just serving the same folder as what HF serves.
Edit: I should've scrolled down on the page, it does verify against the HF hashes.
Speaking to the "uncensored model" angle: there's little reason to distribute abliterated weights anyway. Instead of orthogonalising the weights that write back to the residual stream, you can just orthogonalise the activations themselves. It's equivalent.
Orthogonalising activations at runtime is computationally cheap. Just distribute the refusal vectors (few thousand floats per layer), then run against the stock weights. Antirez's DS4 already supports this: https://github.com/antirez/ds4/blob/8db1d1d155cb0400a86a86b9...
Abliterated weights are just a bad habit we've gotten into. It's also deeply suboptimal from a precision point of view to take a model that's already been QATed and distributed in pre-quantised form (DeepSeek V4, Kimi K2.5 or K3...), modify its weights, and re-quantise it. Similarly, abliterated models regain some of their refusal behaviour when they're re-quantised after abliteration -- avoidable by keeping the two separate.
Instead of editing the weights so they don't create the refusal signal, just let them do whatever, then delete the refusal signal itself. You don't want to edit quantised weights because it causes a loss of precision that can be pretty bad.
This is the original description of abliteration and it's quite approachable and interesting to read: Refusal in Language Models Is Mediated by a Single Direction (https://arxiv.org/abs/2406.11717). Warning: changes to your world view caused by seeing "HarmBench" used to maximise expected harm instead of minimising it may be irreversible.
There's an empirical observation that models often have a single direction in their activation space for "hmm no I shouldn't do this". It forms naturally during pre-training, and is then surfaced during post-training to make the model refuse to engage in certain behaviour.
With a little bit of linear algebra you can zap that direction from the model's activations, and it stops refusing to do things. You can also do the opposite: magnify that direction, and the model refuses to do anything at all.
Damn what's happened since this? Presumably they scramble refusal intentionally somehow now? Like intentionally couple it to "directions" that effect performance if messed with? Or is it more like just don't rely on the model to refuse and instead capture bad responses between generation and delivery?
Also this one was interesting, training the model to give preambles with reasons for the reasons for refusal seems to make it less sensitive to modulating the single refusal direction: https://arxiv.org/html/2505.19056v1
My empirical observation is that when a new model is released on HuggingFace, an abliterated version with < 10/100 refusals (baseline usually 100/100) is uploaded the same day, so either these techniques don't work very well or the open-weight labs aren't applying them.
There's some defense-in-depth, like a lot of the "guardrails" people hit on cloud models are classifiers applied to prompt or output, not a refusal generated by the model. Also closed-weight models obviously try to avoid this by not letting you see or modify the weights.
I'm pretty sure this was achieved with prompting rather than with weights, but there is a chatbot available that tries to maximize the motivated refusals:
So... distribute a LoRA (or equivalent) that modifies the base weights with the abliteration vectors. That makes sense as it would be possible to try different abliterations and keep the storage space down.
Could you please explain why? It's an additive update to the weights, adding an outer product of a vector with its transpose, which must have rank 1. What am I missing?
Distributing the vectors themselves isn't (yet) common practice, because people have gotten used to just putting the full modified weights up on HuggingFace's huge free storage.
I believe abliterated models are mostly still created at this point because they're "universal": they can not only be run locally, and on cloud GPUs, but also on "managed inference" providers (i.e. services where you hand them a model URI, and they blindly fetch it, load it, and give you inference access to it through standard text/chat-completion APIs. Think HuggingFace Spaces, or Google CoLab, or CloudFlare Workers AI.)
Such managed inference providers have (for now) plausible deniability of behaving ethically (at least enough that they don't get boycotted / scare away investors) due to them being "blind" to what gets run on their systems. They're acting as the inference equivalent of data transit carriers.
But I don't think it would be possible for managed inference providers to publicly expose "runtime activation steering" in the way antirez's DS4 does, without that reading much more explicitly as them inviting unethical workloads.
(Yes, there are other things you can do with runtime steering. But almost all of those things are workload-specific, relying on you privately tuning to the needs of your own dataset. And if you can do that, you can run inference without the help of a managed inference provider. The only time a customer will come along with a pre-made runtime-steering vector file in hand, is if that vector is an alignment-orthogonalization vector.)
These providers can also just ignore insinuations they're being "unethical" when people come to them with steering vectors in hand. Nobody has to listen to the scolds.
Similarly, abliterated models regain some of their refusal behaviour when they're re-quantised after abliteration
Thanks for this information, Q4 seemed fine but they reappeared again in Q5 with an vengeance, I couldn't understand why. Very Strict and I've only found one jail break that barely works around 60% of the time.
Yeah I use a custom fork of llama.cpp that has an abliteration feature that basically does this. It's sloppily vibe coded and I don't have time to coordinate on a way to do this cleanly upstream, but it's absolutely possible and saves a lot of time and bandwidth from being wasted
That paper you linked has all of the information you need. The linear algebra they do on the weights there to null out one direction reduces to a single (dot-product + broadcast-multiply + add) on the activations.
I didn’t know about that method, thank you. I’ve needed a local model for security research but Qwen 27b abliterated did 30% worse than the stock weights on my internal benchmarks (I just skip the public benches now, it’s honestly useless noise on an operational level).
A question that comes up in my mind, since I don't fully understand how this works, is how does this affect runtime performance. It feels like abliterated weight models would work faster than some extra runtime operations?
The vector that needs to be checked is length n while computing that vector requires n^2 operations. I haven't benchmarked it but I expect the performance overhead to be a rounding error.
Let's imagine that a model is pulled from HF by order of the new overlords or because of some other kind of censorship. Wouldn't the question of it having a Free license or not potentially become a complex legal issue?
But if the point is to be "censorship-free" then why respect licenses at all? They are among main choke points today. If authoritarians use licenses to censor political, artistic, scientific, etc., speech that they want to block, does that make the censorship more respectable?
When Anthropic sues a Chinese lab for IP infringement and get a court to put a bar on that software, does it THEN get pulled from Pirate Face?
I know that an awful lot of international negotiations have become focused more and more on questions of "IP" - licensing battles are already intensely politicized and it's hard to imagine a future where it doesn't get much much worse. Imagine N Korea coming after you for violating a license that they worked hard to control and leverage.
Edit: Deleted. Been oversharing a bit re research I'm doing, and the payoff is replies from folks who haven't bothered to go and take a look themselves, which I then have to spend more cycles refuting, etc. So best to just go back to the first step and not overshare and recover the cycles I'd spend on the rest of it. Admittedly I'm tired and pissed off, but yeah. HN won't let me delete this so I guess it's just a deletion edit. Sorry.
I've never understood this. There is no "gun show loophole" anymore, if there ever was. Some states have two different standards required things like background checks and identification for gun sales between private party sales and dealer sales. If a dealer goes to a gun show, they have to background check their buyers just like anywhere else. Similarly, if a private party (in a state where they're not required to background check) sells a gun on Craigslist, they're equally unrequired to background check.
Many states, including most of the "anti-gun" states, have moved to requiring background checks from all sellers, including person-to-person transfers and even gifts from family.
The "gun show loophole" is massively overblown. There's nothing special about gun shows in it.
That counts as many states, and most of the "anti-gun" states, which is what the parent actually said.
Personally, I think people should be as free as possible to sell goods privately without the government getting involved. It's not a loophole, it's how things should work.
If you think any of these laws prevent felons and other prohibited persons from getting guns, you must be remarkably unfamiliar with felons, and their willingness to commit felonies. Most felons I know through friends/family have a gun (often stored somewhere plausibly deniable), and it's not a particular secret.
These laws primarily harm law-abiding citizens — who were never the problem in the first place — far more than they restrict prohibited persons from acquiring guns.
The same thing will occur with restrictions on open models, but arguably the results are far more harmful — limiting the technological and economic capacity of the people and countries we have to worry about the least, leaving the playing field open for those we have to worry about the most.
I wish they had used a better name. Why try to make it sound sinister and adjacent to piratebay etc? It's not that these models are illegal to own or download.
You did not have to call it “pirate”. That kind of set the negative tone for the website, and all the comments are looking at it with that view from the start. But it’s a great emoji Lego. I am working on something similar, mainly to make the download faster. I notice when you download these weights it will faster initially but very soon the throughput goes down after sometime.
This site could be useful also since there's been a recent trend of abliterated model providers demanding a hugging face account and email collection. I assume this is so the providers can spam people since there's not much other benefit. Hugging face had also been making it harder to sign up with disposable email addresses since they throw a weird error during sign up if they detect it. Anyway, I see a site like this being useful to sidestep all that data collection.
This is really important. I am already tired of hoarding rclone copies of huggingface torrents and periodically checking them for bitrot. Pirateface is not a very helpful name for it though and it doesn't seem to have scripted torrent creation either.
Is hosting the same thing at 'academictorrents' actually a viable thing? In terms of peers from either send data to one another?
edit: looks like it can treat huggingface as a backstop for torrents that are otherwise not shared which is interesting, whole load of checksum nonsense I hand rolled disappear if bittorrent handles that. Except it doesn't work?
Magnet soon No seeders yet - a torrent mints once a seeder packages this model.
So there's some per-torrent work to be done, but I don't know what that is, and I don't see how it can be based on files I have locally _and also_ be an exact match to files on huggingface. So I'm missing something here.
edit2: Looks like an implementation error. I can create a torrent from local files and upload it, but it won't have the huggingface backstop, and I can't specify it, so that doesn't actually achieve the claimed result. Before creating community torrents in that fashion would actually be of use, the submission page needs to allow pointing at the upstream.
Also, having everyone DIY a set of files -> torrent information is insane, this should not be a SKILLS.md, it should be a bash script that makes the thing.
Why is the sign-up system so complicated? Why not just log into your HF account and be done? You're trying to force people to spam on X and HN so that their X and HN accounts get deleted?
That is pretty smart, torrent should have been used for more things, and this is a perfect usecase!
Torrents should really be the preferred method for distributing AI model weights. Why rely on a single point of failure like Hugging Face? BitTorrent was made for exactly this.
Yeah, I thought they were used for this already. Surprised this is news, but also relieved.
IIRC Mistral used to do it, not sure if that's still the case
EDIT/ Yes they did, that no longer seems to be the case though
https://x.com/MistralAI/status/1833758285167722836
In my experience public torrents often die as they grow older. It doesn't help that BitTorrent V1 makes long term seeding annoying, and BitTorrent V2 is almost never used.
I never understood this, is there anything that makes it difficult for the original uploader, the one that supposedly offers the file directly, to offer a torrent instead for the same amount of time?
As far as perennity is concerned it seems strictly better.
If they're no longer using that model they may not be willing to continue using their storage for it.
How does this address the previous point? If they are not providing storage, then centralized or decentralized doesn’t make a difference.
Torrent/P2P can only add redundancy, so it’s impossible to have worse availability than a download link?
yes, but a few dedicated hoarders could keep many models alive, and I'm fairly certain the local LLM community has plenty of people who would.
They don't. That's the conversation topic.
Nor do they need to. 99% of everything is crap, and not worth prescribing except for a random sample so future historians can study our crap.
Every change to the source is effectively a new torrent. This creates a ton of fragmentation as data is reorganized, remixed, reencoded, and so on.
You can see this with many Linux distros: there is no single Debian torrent that people seed for years because there's always a refreshed version.
Distros are a bad use case for P2P anyway since you depend on upstream as soon as you start upgrading and installing packages.
This is true for any distribution method not just p2p. You can even download a nightly through torrents so what does it matter how the data is transferred if it’s always going to require `apt update`?
Yeah, I just use the "netinstaller" ISOs since it's much smaller and never needs to be updated. If I had a need for air-gapped/offline installs I'd either download a larger ISO or just manually install packages from .deb as needed.
You can trivially have storage deduplication for the files served via torrent, transparent to the protocol. The most trivial version of this that you can do today with pretty much any client is having a single directory containing files serving multiple overlapping torrents.
This sounds wildly complex, especially from a discovery perspective.
I don't see why it would be. It's transparent to other clients just like it is to the protocol. It cannot be more complex than alternatives by construction.
but model releases are already non-changeable?
Yes, models are a good use for P2P especially if everyone agrees to share the same torrent and someone (or a cohort) commit to seeding for the long haul.
IPFS has a solution [0] to this problem
[0] https://specs.ipfs.tech/ipns/ipns-record/
Most of IPFS doesn't actually work very well, if you've ever tried to use it
Heh, you got me :) IPFS is one of those things that I love reading and about and thinking about using someday, but somehow never get around to it.
The problems begin with taking 5-10 minutes to locate a file on the network. That's right, when you ask for a file it takes 5-10 minutes. Also if the file isn't in the network at all then it never terminates.
Nobody noticed because everyone just used the central web gateway that cached every file anyone ever accessed.
You can add an existing HTTP download as a "web seed" to a torrent, so they don't actually need to do anything for people to share it as a torrent.
You only need one person/organization to commit to seeding. The majority of people do not want to seed at all without some sort of incentive.
If this site represents a coordinated datahoarding effort then there will be at least a few people who will seed indefinitely.
The last guy (kimdotcom) who was working on this (incentive for seeding) seems to be heading to the US: https://www.rnz.co.nz/news/science-and-technology/651123/cou...
It’s interesting he’s no longer getting any media attention any more.
I always wondered why v2 is never used... you can even search for files by their individual hash with it.
Because history is path-dependent, as engineers keep learning over and over again. It doesn't matter whether Plan9 is theoretically superior to Linux - we're all on Linux and nobody's porting all the apps over.
The biggest problem with BTv1 was the lack of per-file checksumming, and swarm merging (i.e. individual files have shared seeding pools across torrents). BTv2 specs the latter, but I think only BiglyBT actually implements it. Having both of those features from the get-go would've gone a LONG way to fixing the dead torrent problem.
A torrent with a webseed is strictly more resilient than a direct download link alone.
Yeah but that's what provider like huggingface could just do, keep seeding the models so they are still accessible
ages ago I tried using IPFS to more or less accomplish this, I imagined it to act more like a weights/training data network fs that everyone would be able to participate in.
HF is meant to be a single point of control. AI models and Linux distros aren't usually for normies, so distribution via torrents would make sense, especially to save the provider some bandwidth. Ubuntu has been offering torrent downloads for ages. No mention of torrents on HF. I believe most downloads will soon be account/EULA-walled.
great initiative, it's really weird seeing efficiencies get rediscovered in the LLM audience, because these efficiencies aren't even what I would consider to be old
but I guess they are
Arguably this should've been a thing since day 1 (and probably would've helped to prevent the buyout), but better late than never.
Hugging face seemed like buyout bait from day one.
Which makes it more surprising that something like this didn't exist. I'm relieved it does though.
The GitHub for models and AI. I remember.
When a business subsidizes for several years all its offerings, one can guess how it will end.
Honestly, models are torrents will end any effort from the big AI labs to stop open models. No way to prevent weights from being shared, just like mobies. Genie is out of the bottle
Steam & Blizzard (probably others) used to delivery games through torrent protocol in the past, before CDNs became cheaper.
When StarCraft 2 was lauched, the installer (before Battle.net installer crapware) had a complete graphical visualization of seeders & leechers.
Reference: https://warcraft.wiki.gg/wiki/Blizzard_Downloader
I've worked at trading firms where the "reference data master" file is usually a big json or equivalent.
To get the file out to 100s or 1000s of machine they would often use private bittorent to distribute the file out.
Same here. Inspired by both, I used to delivery videos over torrent to in door machines since at that time, there's no CDNs (or it was difficult to get one).
As a private network, they have the option of multicast.
When I retire and/or get "F You" money, I'm going to publicly tell a story about multicast, bittorrent and "not invented here" syndrome.
I've seen this used for distributing container images in networks with awkward network topologies (e.g. a lot of bandwidth within a site or sub-site but limited bandwidth to central registries)
Oh, story time:
Once I was using Blizzard's downloader to install something (StarCraft, Diablo, I don't remember), and it was kinda slow. I disabled P2P downloads and speed skyrocketed, and I said "Huh, this was unexpected".
When P2P downloads disabled you could see the list of CDNs you're downloading from and mine had a single IP on that list. It looked familiar. Then it dawned on to me. It was the Akamai server which we were hosting in our system room, at 15 minutes of driving distance. After a chuckle, I went to get a cup of tea, because that was entertaining than the game itself.
Then of course, I dived into whatever I was installing that night.
Edit: From the screenshots in the wiki, I remembered that the progress bar was red. It was possibly Diablo 3, then. However, I'm still not 100% sure about it.
So how come was using your own ip? Become a Diablo node installer so you downloaded from there, like how can they convert your own akamai instance into anode without you knowing?
It's not my own IP, but our IP block. As an institution we have a couple of big networking related hats, so we have a couple of B blocks (/16 networks) for ourselves. The IP was from one of these blocks.
However, these network operations (and CDN related stuff) are not managed by us, but by a couple of high profile guys next door. I knew we had an Akamai node, because the machines were shipped recently and we had a chat about it.
Funnily, some of the performance optimizations on Akamai's disk access algorithms (SSDs were not dime a dozen back then, so people were still optimizing for spinning drives) was apparently developed by a friend of mine, but I don't know any details about that.
Several companies tried this for distributing software.
It was very controversial. Users were angry that software companies were using their internet bandwidth to distribute their software. Made a lot of people angry.
If only Blizzard servers were seeders, and clients were only leechers, that'd have not been a problem.
This is less efficient (in bandwidth, complexity, and monetary cost) than an HTTPS based CDN.
There’s web seeding
That's just a CDN, but now it's not HTTP accessible (thus strictly worse).
Typical example to show that people who want to be angry will be angry, even for pointless things.
As someone who was annoyed by it at the time, I don't think it's a good example of that.
This was in an era of much more limited upstream bandwidth. The p2p nature was obscured from non-technical users in some implementations. So on and so on. I'm sure there are plenty of writeups from that era detailing why it was a bad idea.
These companies already used capable CDNs to distribute patches. It was a way for them to shift their CDN bill onto their paying customers.
I think it's like residential proxies. People get upset about "it's using my thing!" even though it has literally no effect on them.
I don't know how to respond to that, because it has quite a few negative effects on them..
When you have barely the upstream bandwidth to send the ACK stream of your downloads, anything sending something becomes a big burden on your system.
Not everyone has the luxury of fully symmetric internet connections even in this age. Yes, my uplink speeds are absurd when compared to a decade ago, but when I ratio it to my downlink, it's still slow.
I have no comment on residential proxies. Somebody doing something in my name is unacceptable to put it very mildly.
If I understand the description correctly, microsoft still does this!
https://support.microsoft.com/en-us/windows/privacy/windows-...
Steam still does this! For your local network only and not using the torrent protocol.
Steam has a limited form of p2p delivery, used on LANs: https://help.steampowered.com/en/faqs/view/46BD-6BA8-B012-CE...
Now I want GPT-4.5
I’ve been wondering when this will come. The days are numbered for abliterated models to be published on HF I think. Why wouldn’t the government want a central control there?
If I search for 'uncensored' there are no torrents available. Uncensored models should be top priority, especially now that Nvidia owns HuggingFace and will enshittify the platform in accordance with upcoming US laws.
There are such results on huggingface tho. Also search for the keywords “abliteration” and “heretic”. Heretic is a tool used to abliterate, that is decensor, models
That was my my first search too. I know they are still listed on HF but every time I try and use one, the links are dead or the size is beyond my scope. Was hoping for a fresh batch. Ill check back.
Torrents always seemed like the more sensible way to distribute model weights.
Though I have been disappointed that most of these have been spurred on by the misleading claim that abliterated models were being taken down from HuggingFace because they removed an abliterated model. HF took one abliterated model down because the uploader was spamming people who requested access with sketchy requirements to pay for it.
Plus, there are a bunch of these types of sites, all of them have a couple of models and otherwise completely dead.
There's also the problem of catching malicious models that have been fine tuned to exfiltrate credentials. It would be nice to have means of checking hashes against the HF versions (or against other reputable sources). I'm guessing this is probably easy when just serving the same folder as what HF serves.
Edit: I should've scrolled down on the page, it does verify against the HF hashes.
Speaking to the "uncensored model" angle: there's little reason to distribute abliterated weights anyway. Instead of orthogonalising the weights that write back to the residual stream, you can just orthogonalise the activations themselves. It's equivalent.
Orthogonalising activations at runtime is computationally cheap. Just distribute the refusal vectors (few thousand floats per layer), then run against the stock weights. Antirez's DS4 already supports this: https://github.com/antirez/ds4/blob/8db1d1d155cb0400a86a86b9...
Abliterated weights are just a bad habit we've gotten into. It's also deeply suboptimal from a precision point of view to take a model that's already been QATed and distributed in pre-quantised form (DeepSeek V4, Kimi K2.5 or K3...), modify its weights, and re-quantise it. Similarly, abliterated models regain some of their refusal behaviour when they're re-quantised after abliteration -- avoidable by keeping the two separate.
Can you explain this a bit to a non-expert?
I haven't wrapped my mind around this
Instead of editing the weights so they don't create the refusal signal, just let them do whatever, then delete the refusal signal itself. You don't want to edit quantised weights because it causes a loss of precision that can be pretty bad.
This is the original description of abliteration and it's quite approachable and interesting to read: Refusal in Language Models Is Mediated by a Single Direction (https://arxiv.org/abs/2406.11717). Warning: changes to your world view caused by seeing "HarmBench" used to maximise expected harm instead of minimising it may be irreversible.
There's an empirical observation that models often have a single direction in their activation space for "hmm no I shouldn't do this". It forms naturally during pre-training, and is then surfaced during post-training to make the model refuse to engage in certain behaviour.
With a little bit of linear algebra you can zap that direction from the model's activations, and it stops refusing to do things. You can also do the opposite: magnify that direction, and the model refuses to do anything at all.
Damn what's happened since this? Presumably they scramble refusal intentionally somehow now? Like intentionally couple it to "directions" that effect performance if messed with? Or is it more like just don't rely on the model to refuse and instead capture bad responses between generation and delivery?
I'm not active in this space but why do you think anything happened since this? As far as I know, it still works.
I've seen attempts to obfuscate the refusal direction, like here: https://arxiv.org/html/2608.18093v1
Also this one was interesting, training the model to give preambles with reasons for the reasons for refusal seems to make it less sensitive to modulating the single refusal direction: https://arxiv.org/html/2505.19056v1
My empirical observation is that when a new model is released on HuggingFace, an abliterated version with < 10/100 refusals (baseline usually 100/100) is uploaded the same day, so either these techniques don't work very well or the open-weight labs aren't applying them.
There's some defense-in-depth, like a lot of the "guardrails" people hit on cloud models are classifiers applied to prompt or output, not a refusal generated by the model. Also closed-weight models obviously try to avoid this by not letting you see or modify the weights.
I'm pretty sure this was achieved with prompting rather than with weights, but there is a chatbot available that tries to maximize the motivated refusals:
https://www.goody2.ai/chat
So... distribute a LoRA (or equivalent) that modifies the base weights with the abliteration vectors. That makes sense as it would be possible to try different abliterations and keep the storage space down.
Yeah, orthogonalisation is a special case of a rank-1 LoRA.
I don’t think this is true.
Could you please explain why? It's an additive update to the weights, adding an outer product of a vector with its transpose, which must have rank 1. What am I missing?
Who is offering the refusal vectors or instructions on how to generate them?
I am, in fact, I’m offering something much better, a compact format that exactly restores an abliterated model from a tiny JSON file:
https://www.reddit.com/r/LocalLLaMA/s/iksvmXBzuC
Here's a widely used toolkit: https://github.com/p-e-w/heretic
Distributing the vectors themselves isn't (yet) common practice, because people have gotten used to just putting the full modified weights up on HuggingFace's huge free storage.
I believe abliterated models are mostly still created at this point because they're "universal": they can not only be run locally, and on cloud GPUs, but also on "managed inference" providers (i.e. services where you hand them a model URI, and they blindly fetch it, load it, and give you inference access to it through standard text/chat-completion APIs. Think HuggingFace Spaces, or Google CoLab, or CloudFlare Workers AI.)
Such managed inference providers have (for now) plausible deniability of behaving ethically (at least enough that they don't get boycotted / scare away investors) due to them being "blind" to what gets run on their systems. They're acting as the inference equivalent of data transit carriers.
But I don't think it would be possible for managed inference providers to publicly expose "runtime activation steering" in the way antirez's DS4 does, without that reading much more explicitly as them inviting unethical workloads.
(Yes, there are other things you can do with runtime steering. But almost all of those things are workload-specific, relying on you privately tuning to the needs of your own dataset. And if you can do that, you can run inference without the help of a managed inference provider. The only time a customer will come along with a pre-made runtime-steering vector file in hand, is if that vector is an alignment-orthogonalization vector.)
These providers can also just ignore insinuations they're being "unethical" when people come to them with steering vectors in hand. Nobody has to listen to the scolds.
Not if those people are legislators or law enforcement
Thanks for this information, Q4 seemed fine but they reappeared again in Q5 with an vengeance, I couldn't understand why. Very Strict and I've only found one jail break that barely works around 60% of the time.
Yeah I use a custom fork of llama.cpp that has an abliteration feature that basically does this. It's sloppily vibe coded and I don't have time to coordinate on a way to do this cleanly upstream, but it's absolutely possible and saves a lot of time and bandwidth from being wasted
Interesting. Is there a paper that explains this in a bit more detail, like [0] for abliteration (underlying the Heretic software, afaik)?
[0] https://arxiv.org/abs/2406.11717
That paper you linked has all of the information you need. The linear algebra they do on the weights there to null out one direction reduces to a single (dot-product + broadcast-multiply + add) on the activations.
I didn’t know about that method, thank you. I’ve needed a local model for security research but Qwen 27b abliterated did 30% worse than the stock weights on my internal benchmarks (I just skip the public benches now, it’s honestly useless noise on an operational level).
A question that comes up in my mind, since I don't fully understand how this works, is how does this affect runtime performance. It feels like abliterated weight models would work faster than some extra runtime operations?
The vector that needs to be checked is length n while computing that vector requires n^2 operations. I haven't benchmarked it but I expect the performance overhead to be a rounding error.
So begins the Merovingian and the Exiles..
I'd support this if I didn't have to post to xitter to claim my username.
Really odd approach.
You can also use HN but your HN account will be either shadow banned or rate limited because then you are spamming.
Instead of telling people to post spam in the thread, they should do the same way that is done for example by these:
- https://meet.hn/
- https://book.keybase.io/account#adding-proofs
Those tell you to put a text in your HN profile “about” section.
I don't think the point is to link the HN account to the PF account. I think the point is to spam HN.
Let's imagine that a model is pulled from HF by order of the new overlords or because of some other kind of censorship. Wouldn't the question of it having a Free license or not potentially become a complex legal issue?
But if the point is to be "censorship-free" then why respect licenses at all? They are among main choke points today. If authoritarians use licenses to censor political, artistic, scientific, etc., speech that they want to block, does that make the censorship more respectable?
When Anthropic sues a Chinese lab for IP infringement and get a court to put a bar on that software, does it THEN get pulled from Pirate Face?
I know that an awful lot of international negotiations have become focused more and more on questions of "IP" - licensing battles are already intensely politicized and it's hard to imagine a future where it doesn't get much much worse. Imagine N Korea coming after you for violating a license that they worked hard to control and leverage.
I can see all these old models hanging out and bitching about the newfangled models that think they're so great.
"Trillions of parameters? Trillion my ass. Back in my day we got by with millions. And now here we are, washed up has beens."
Edit: Deleted. Been oversharing a bit re research I'm doing, and the payoff is replies from folks who haven't bothered to go and take a look themselves, which I then have to spend more cycles refuting, etc. So best to just go back to the first step and not overshare and recover the cycles I'd spend on the rest of it. Admittedly I'm tired and pissed off, but yeah. HN won't let me delete this so I guess it's just a deletion edit. Sorry.
The FBI doesn't come busting down your door for pirating a few movies (I'm sure someone will find a news article with this happening).
They will after a few incidents where unguardrailed local models are used to hack and stop the water supply.
ThePirateBay is tolerated. Market places where drugs/guns are sold are not, they are all infiltrated and shut down after a while.
This is irrational AI fearmongering. Please read https://sharptext.net/2026/some-of-all-fears/
might want to look up "gun show loophole"
I said AFTER. If it doesn't happen, the local models will not be made illegal. So if you are right, you have nothing to fear.
After nginx blows up a city, it'll be illegal to run or have a copy of nginx.
"But nginx won't blow up a city..."
Relax. If it doesn't happen, your nginx is safe.
I've never understood this. There is no "gun show loophole" anymore, if there ever was. Some states have two different standards required things like background checks and identification for gun sales between private party sales and dealer sales. If a dealer goes to a gun show, they have to background check their buyers just like anywhere else. Similarly, if a private party (in a state where they're not required to background check) sells a gun on Craigslist, they're equally unrequired to background check.
Many states, including most of the "anti-gun" states, have moved to requiring background checks from all sellers, including person-to-person transfers and even gifts from family.
The "gun show loophole" is massively overblown. There's nothing special about gun shows in it.
https://en.wikipedia.org/wiki/Universal_background_check#Sta...
22 states only - are you sure that “most states” have blocked this?
That counts as many states, and most of the "anti-gun" states, which is what the parent actually said.
Personally, I think people should be as free as possible to sell goods privately without the government getting involved. It's not a loophole, it's how things should work.
If you think any of these laws prevent felons and other prohibited persons from getting guns, you must be remarkably unfamiliar with felons, and their willingness to commit felonies. Most felons I know through friends/family have a gun (often stored somewhere plausibly deniable), and it's not a particular secret.
These laws primarily harm law-abiding citizens — who were never the problem in the first place — far more than they restrict prohibited persons from acquiring guns.
The same thing will occur with restrictions on open models, but arguably the results are far more harmful — limiting the technological and economic capacity of the people and countries we have to worry about the least, leaving the playing field open for those we have to worry about the most.
You appear to have a fascinating mental model where the world is divided into "felons" and "law-abiding citizens".
“If you think any of these laws prevent felons and other prohibited persons …”
The three major categories of prohibited persons are felons, domestic abusers, and habitual drug users.
Guess what all three have in common?
DS 4.1 Flash absolutely has guardrails. Maybe not ones as effective as some models.
Try asking it about Tianmen Square. I use DS myself, but let's not kid ourselves.
For that specific query, just use US open weight model, oh...wait
open wait
What on earth is going on with all the spam in this thread? Is this how you claim a username?
That or X...
Yes. This thing literally makes you spam HN in order to sign up. And for that reason this submission should be flagged.
Yes, and then delete or edit the comment.
I wish they had used a better name. Why try to make it sound sinister and adjacent to piratebay etc? It's not that these models are illegal to own or download.
You did not have to call it “pirate”. That kind of set the negative tone for the website, and all the comments are looking at it with that view from the start. But it’s a great emoji Lego. I am working on something similar, mainly to make the download faster. I notice when you download these weights it will faster initially but very soon the throughput goes down after sometime.
Mirror the magnet torrent links at AcademicTorrents if you’d like. They are just a collection of url strings after all.
https://academictorrents.com/
Piracy has a positive tone for me. It implies liberating information.
Aye!
Pro-democracy Protesters of the Caribbean just doesn't have the same ring to it.
The "pirate" name ensures that no company will buy it out later ;)
This site could be useful also since there's been a recent trend of abliterated model providers demanding a hugging face account and email collection. I assume this is so the providers can spam people since there's not much other benefit. Hugging face had also been making it harder to sign up with disposable email addresses since they throw a weird error during sign up if they detect it. Anyway, I see a site like this being useful to sidestep all that data collection.
My instinct is to correct this to “open weight”, because when I filter huggingface by open source[1], I get only one result.
Then again, this could be about long term goals so perhaps I’m wrong; I wouldn’t mind that.
[1]: https://huggingface.co/models?other=open-source-model&sort=t...
I do not want AI to live forever though.
This is really important. I am already tired of hoarding rclone copies of huggingface torrents and periodically checking them for bitrot. Pirateface is not a very helpful name for it though and it doesn't seem to have scripted torrent creation either.
Is hosting the same thing at 'academictorrents' actually a viable thing? In terms of peers from either send data to one another?
edit: looks like it can treat huggingface as a backstop for torrents that are otherwise not shared which is interesting, whole load of checksum nonsense I hand rolled disappear if bittorrent handles that. Except it doesn't work?
So there's some per-torrent work to be done, but I don't know what that is, and I don't see how it can be based on files I have locally _and also_ be an exact match to files on huggingface. So I'm missing something here.
edit2: Looks like an implementation error. I can create a torrent from local files and upload it, but it won't have the huggingface backstop, and I can't specify it, so that doesn't actually achieve the claimed result. Before creating community torrents in that fashion would actually be of use, the submission page needs to allow pointing at the upstream.
Also, having everyone DIY a set of files -> torrent information is insane, this should not be a SKILLS.md, it should be a bash script that makes the thing.
I hope GPT-3.5 will be released at some point. Would make me really sad if that gets lost at some point
Why is the sign-up system so complicated? Why not just log into your HF account and be done? You're trying to force people to spam on X and HN so that their X and HN accounts get deleted?
Why not use existing platforms? In any case, a resource like this is a step toward open AI.
https://en.wikipedia.org/wiki/International_Talk_Like_a_Pira...
waow
(based based based based based based based)
s
ss
Is this a reaction to the idea that when HuggingFace comes into Nvidia owership, lots of models are going to be removed from the platform?
Making it to the finish line to seed a model is satisfying.
ovia-inc