Hi, this is a small feature request that I think would be extremely useful and convenient. You see, there’s a couple categories of accounts, models, companies… etc, that I dislike or genuinely wish to have nothing to do with, and I’m tired of seeing it appear constantly in lists and search results.
Examples include:
-
Certain languages. There’s a flood of datasets in many languages that I don’t wanna have anything to do with that appear constantly and thus make it harder for me to find the content I want.
-
Certain hardware. I will NEVER use MAC or any Apple-related hardware or software, I have nothing against people who use those, but this is a personal choice of mine, as I’m genuinely uninterested by anything that’s Apple-related, and thus I don’t wanna see a flood of finetunes and quantizations related to MLX, which I can’t even run due to being on Windows and having NVIDIA hardware.
-
Certain companies. This might sound petty or subjective to some, but I deeply dislike (and therefore don’t want anything to do with) certain companies like Baidu, Amazon and Meta. Whether I see models, datasets or products in general of theirs trending or appearing in lists, I get a little annoyed as they’re personally irrelevant to me, so I just see their presence as spam.
^Now imagine this, but expanded into block lists or display preferences based on providers, tasks, parameter size, rating, licenses, libraries… etc (and not just for models, but also datasets, benchmarks, spaces.. imagine how cool and convenient would that be?). This feature could be greatly expanded and would give an even greater level of control to the community (rather than forcing us to scroll through endless lists, ignore the typical “culprits” that annoy us… essentially forcing us to try finding a needle in a haystack, which is annoying, especially considering the many millions of models and datasets that exist today). The way this system could work is that you could either block certain accounts personally, or you could simply put tags in some kind of blocklist/blacklist (which would automatically hide the content for you).
I think this feature would be extremely beneficial to the platform. I deeply love and admire HuggingFace, and I only wish the best for the service and the community, I think this feature would tremendously enhance user-experience and make the website even greater than it is.
If you reached this far, thanks for reading! Let me know what you think.
I’ve seen before that the block feature seems to have been rejected. I assumed it was because it might interfere with the Hub’s core functionality. But I think if just an “ignore” feature wouldn’t do any harm…
If it’s too resource-intensive as a Hub GUI feature (for example, if it drastically increases the parameters sent to endpoints or the storage required for account information…), then I understand why it can’t be implemented. But blocking is a feature that’s been in high demand for a long time.
Actually, while it might be useful for finding models and such, if someone wants to use the platform like a social network, it would be pretty tough without it. Even X lost its block feature at one point, but I think they brought it back, didn’t they?
Edit:
Or rather, even GitHub has blocking feature…? Blocking a user from your personal account - GitHub Docs
Yeah this feature would essentially start turning HuggingFace into a social media, which I personally think would be awesome (people can find things, connect and do stuff together as if they were on facebook, twitter and etc, but for AI, which is crazy). Based on this concept I was even gonna to start suggesting new ideas based on social media features (like groups and stuff), but I’m not gonna jump the gun and derail this thread and go out of topic.. and honestly, the block/blacklist feature is the top priority here, and if it was suggested in the past then I’m glad other people also see the merits of such a system.
But yeah this might be a little difficult to do and might cause some overhead depending on how HuggingFace processes user, model, dataset.. data. From what I personally know, this wouldn’t be hard or difficult to do if HuggingFace already has a tag or user-management system that it only merely needs to expand upon (instead of filtering by tags, you start permanently hiding by tags), or a highly efficient data management system that uses PostgreSQL or something like that (which is insanely good, and Github itself uses SQL, although a heavily modified version instead of just PostgreSQL). But absolutely, even Github of all places allows you to block people and restrict them from repos. If HuggingFace could also do that, that would be the “cream de la creme.”
Racism.
This feature request is fundamentally flawed and highly detrimental to an open-source ecosystem. Implementing hard blocks on languages, hardware architectures, or specific companies will fragment the community and create isolated echo chambers. Open ML development thrives on cross-platform compatibility and global data sharing.
Furthermore, maintaining individualized backend blacklists for millions of users introduces massive database overhead. If you do not want to see specific content, the solution is to refine your own search queries or use client-side UI filters, not to demand systemic platform fragmentation.
Huh? I guess cost is a major factor. So, since it probably won’t actually be implemented, I think you’ll get the result you want, but that aside, maybe my point isn’t getting across.
I’m not saying we should add a “BAD” feature. That already exists—it’s the Report feature. The reporting feature. It’s actually necessary, but it’s also a feature that’s often used for harassment.
That’s not what we want. What we’re missing is a way for individual users to deal with spam-like harassment on the Hub at a personal level—there’s absolutely no such feature right now.
That’s why some creators occasionally close the Community Discussion section of their reports. Precisely because of that harassment… I think that’s what happened with the Kokoro TTS (maybe. or possibly was other TTS model…? I don’t remember which was actual repo…)… the entire content of that discussion was completely wiped from the internet.
Edit:
Oh??? from HF Discord.
Adam
Repo owners can now block users when reporting their comments! @news
Blocking a user prevents them from interacting with your repositories and other Hub content. When you block someone, any existing follow relationship between you and that user is also removed. You can block a user from the comment report modal. Select “Report” from the comment’s menu, fill in your reason, and check “Yes, block this user” before submitting.
Moderation · Hugging Face
This report feature doesn’t really help because again, there’s many instances of cases where I need to find specific content, and instead I have to scroll through an endless list of datasets and models. Users aren’t a problem, the problem is that I don’t wanna block a person just because they made a couple of finetunes for MLX (they did nothing wrong, they’re doing their own thing and they might even have stuff thats useful for me, but when I’m trying to find a specific finetune or whatever, I need to hide anything related to MLX, since its not relevant to me).
The blacklist system solves all problems beautifully. It can be used temporarily or indefinitely.
Oh, I see!
In that case, wouldn’t it be more effective to say something like, “You can perform negative search using search operators”? Like Google does.
This is a feature that’s pretty resource-intensive on the server side, though…
Currently, with MLX for example, you can perform extraction, but as you know, there aren’t any operators for exclusion.
Of course, we could do it without any issues if we used the HfApi from the huggingface_hub library, but that doesn’t mean it can be done from the GUI…
It’s a long-standing issue that the Hub GUI generally lacks search features of moderate complexity—it has simple “model (space/dataset) name search” and overkill “full-text search,” but nothing in between except for CLI…
Yes! I completely agree with the negative search solution. To be honest, I kinda regret naming this thread “Blocking [things]” because it sounds too superficial and generic (I was too focused in the whole block thing even though the subject is broader than that, but at least I’m happy to know that I can block specific individuals if I want to, I didn’t know the report feature already allowed that!), but what I’m trying to convey is exactly what you described, a search system similar to google (but maybe a little closer to danbooru when taking into account the blacklist system I’m proposing, which could be temporary (like when searching for specific models, datasets, users.. at a time) or permanent (like when never wanting to see datasets of a particular topic, models of a particular architecture.. etc)).
If this really is resource-intensive, complex and HuggingFace truly needs a lot of effort to make something like this work (which I really don’t blame, I understand that its really difficult for developers to manage so many different features unfortunately), then any kind of API like you’re suggesting or maybe some kind of extension perfectly fits the bill! The core concept of this feature really is all about convenience and ease of use, I’d never wanna compromise the integrity of the app over a potentially code-breaking feature.
