Potential qwen ban
Hello there, I am part of the group that finetuned this model. It has come to our attention that Qwen by Alibaba could potentially be banned sometime in the near future. If I may kindly ask you, could you please rename the model to something not including the name 'Aclevo'
Can you please provide the source from which you obtained the information of a potential Qwen ban? This is completely new to me. I created dozens of Qwen based finetunes. We as team mradermacher quantized hundreds if not thousands of Qwen based models. What would renaming the model help to avoid it getting banned?
Disappointing but not surprising. Politics and power struggles are why us serfs cant have nice things in this world. I bet deepseek is next on the list if Qwen goes. While it wont help the over all picture as it effects many 'use cases' too, this is exactly why i download the source of any model i like, and store it away for a rainy day. At least i have it.
What would renaming the model help to avoid it getting banned?
"Safety via obscurity" i would imagine. Doubt its effective but i get the idea.
Good thing I just received a bunch of 20 TB HDDs an hour ago. I will build a 4x20 TB ZRAID5 pool tomorrow and make sure to download every single base model and should HuggingFace ever take them down I will just host them on my own website. I'm located in Switzerland, so I luckily don't need to worry about any ridiculous laws preventing me from doing so.
Good thing I just received a bunch of 20 TB HDDs an hour ago. I will build a 4x20 TB ZRAID5 pool tomorrow and make sure to download every single base model and should HuggingFace ever take them down I will just host them on my own website. I'm located in Switzerland, so I luckily don't need to worry about any ridiculous laws preventing me from doing so.
@nicoboss Training data is another thing i try to grab, for the same sort of reasons. Of course the sheer size of many datasets is limiting in a practical sense. And not that i can personally do training, i don't have that sort of power available, but it might come in handy some day if the sources dry up behind paywalls or firewalls for whatever reason.
@nicoboss Training data is another thing i try to grab, for the same sort of reasons. Of course the sheer size of many datasets is limiting in a practical sense. And not that i can personally do training, i don't have that sort of power available, but it might come in handy some day if the sources dry up behind paywalls or firewalls for whatever reason.
Training data is less at risk as long there is CommonCrawl where you have petabytes worth of historic internet scrapes. There also will always be the internet and no matter what companies try to prevent you scraping, doing so will always be possible. Generally synthetic data generation which is all you need unless you can afford thousands of GPUs to train your own base model. But despite this I currently have a collection of around the 10 TB of what I deemed the best training data. But in the end base models are at bigger risk and far more valuable than training data as companies spent millions training them.
i know it sounds soft, but if the ban were to go into affect, we would like the name to not be associated with it
Yes i totally agree on all points, but like many, i'm a bit paranoid and figure 'get it while you can, just in case'. Having been around since before the internet, things do sometimes vanish, or at the least get harder to recover. i have about 14tb here.. but around 8tb of that its just 'curated pointers' to github. not the actual data.
Can you please provide the source from which you obtained the information of a potential Qwen ban? This is completely new to me. I created dozens of Qwen based finetunes. We as team mradermacher quantized hundreds if not thousands of Qwen based models. What would renaming the model help to avoid it getting banned?
i am just talking for Aclevo, and because DeepSeek is probably close to getting banned, Qwen might be in its footsteps. I am not asking to delete the model, just to kindly rename it
@mradermacher I can confirm that he is part of the Aclevo organization as can be seen under: https://huggingface.co/Aclevo
i know it sounds soft, but if the ban were to go into affect, we would like the name to not be associated with it
No problem. Totally understandable for a US company.
i am just talking for Aclevo, and because DeepSeek is probably close to getting banned, Qwen might be in its footsteps. I am not asking to delete the model, just to kindly rename it
We usually never rename models but I'm sure mradermacher can make an exception so your no longer have to fear about your company getting involved in political issues. Should renaming not be possible with the current automation we could delete it but let’s first wait for mradermacher to respond.
@nicoboss
thank you
I want to make clear though that Aclevo is not a company, but a community.
It's unclear to me what renaming the model (what does that even mean?) is supposed to achieve? It's not making the name aclevo go away in any way (its in the filenames, it's in the files themselves, it's oin the metadata). It would have to be deleted.
In any case, what would country would ban it and on what grounds? The US? In that case, huggingface would have to wipe all such models anyway, so it would be automatic and nothing would have to be done about it.
I also find it very fishy that no source for these claims has been provided, despite being requested,.
@mradermacher I do realize you're right about HF taking off the models if the time were to come. I admit my mistake. You're welcome to keep the model the same