Search found 15 matches

This would be useful to archive models such as Wav2Vec, VoxPopuli, Whisper. These models are state-of-the-art for transcribing audio.

Hi, since Google is publishing models (Gemma2), could we get a Google section under Large Language Models please?
NVIDIA and Apple are publishing models also.
Edit: under LLM/Others is fine
!1st

+1, currently most uploads give links to hugging face and specify model name, that's all.
Quality of releases is important

Perhaps there could be some kind of standardized format for SD model uploads? Including the original link, such as to civitai if it exists there, would be good. But if that's down or gone, then also including key information about the model could be important, like recommended schedulers, CFG, prompt keywords, etc.
It'll be better to implement these kinds of rules sooner rather than later, so not many uploads need to be changed. This could a real alternative since civitai has been refusing to use torrents.
Another thing to consider is an RSS feed, just in case there are any seedbox or data hoarding chads out there who want to automatically grab and seed everything. I would certainly be interested in doing this.
Edit: I just noticed there is some sort of RSS type feed, but the URL doesn't seem to work properly with qBittorrent or autobrr. And there doesn't seem to be feeds for the larger sections, such as Image Diffusion Models.

+1

With the recent reports of malware being injected into AI tools, https://www.404media.co/hackers-target-ai-users-with-malicious-stable-diffusion-tool-on-github/, I think making a VirusTotal scan mandatory would be helpful to combat this. While it has some limitations, for example some types of malware are not detected, it should help with blatant and known samples. So how this would work is scan files using VT before upload and include links in your post (also provides file hash, which may be useful for confirming the files source.

Grumpy indeed!

You can upload models that make nsfw, just don't post images

"No posting of NSFW (Not Safe For Work) images or content on the forums."
IMO this is a very weird rule for a site that wants to facilitate access to, and counteract removals and censorship of generative models.
A NSFW rule is not required and counterproductive; i.e. I won't upload anything with that rule in place. Please remove the rule, so I can upload the defecation lora.
BTW, when making a post it says "Notify me when on replies", which makes no sense. Or maybe "replies" is slang for some drug?
EDIT: I also don't want that avatar. Or any avatar. Can't delete it.

I don't really want to add too many categories. Most are downloading a quant of some sort and you can use the search to do something like searching "q4*" in LLaMA 3 8B to find all 4bit quants of models based on L3 8B

Currently all the torrents are kinda just all over the place, which could be a problem if more models are added. I think a separate Quantization section should be added, to dump all the gguf/awq/gptq/exl2 quants and those can be linked to from a single post, one for each model. This way they can be searched but also don't take up space on the model sections.

Think you can only use emojis in posts and PMs

Thanks; one thing though is that trying to use any emojis in my profile section causes a db error, are they intentionally not allowed?

Done. Didn't know if there would be any demand for that, but I guess there're no traditional style discussion forums for ai stuff

I think a section for LLM discussion would be nice here