A minute of silence for wizardlm 2 (gpt-4) about wizardlm HOT 5 CLOSED

Okay, but is there any clarification of what actually happened? I am hearing a bunch of crazy rumors like it's on par with GPT-4 which I highly doubt, and so they removed it for that reason. Open source models in my opinion haven't even surpassed GPT 3.5 if you for example ask them for translations or more obscure facts, they only are better at reasoning and more simple common questions or maybe more focused models with coding etc. I just have a hard time beliving they turned what I would personally call for my use cases a sub GPT 3.5 model into a GPT 4 tier one, but I am downloading it now (mirrored). I highly doubt the model is suddenly good at these things but the cat is already out of the bag and already found a new owner if true

from wizardlm.

Yeah already used a English word for no reason 💔

Not even close to GPT4 at least for my uses, going to try other things I guess. I think Japanese might not be fair as it only has 3200 vocab and is likely not made for this but does illustrate my point well. GPT 3.5 doesn't have this issue btw

from wizardlm.

Update: https://twitter.com/WizardLM_AI/status/1780101465950105775

[WizardLM](https://twitter.com/WizardLM_AI)
[@WizardLM_AI](https://twitter.com/WizardLM_AI)
🫡 We are sorry for that.

It’s been a while since we’ve released a model months ago😅, so we’re unfamiliar with the new release process now: We accidentally missed an item required in the model release process - toxicity testing. 

We are currently completing this test quickly and then will re-release our model as soon as possible. 🏇

❤️Do not worry, thanks for your kindly caring and understanding.

from wizardlm.

Holy crap this model is good, just don't ask it to translate anything I guess. I have never seen a open source LLM get this right, so I think it might actually replace GPT 3.5 for me ignoring speed and Japanese. I am seriously impressed every other model failed this miserably but this is as good as GPT 3.5 and I would say it formatted it better than GPT 3.5

Not scientific at all but open source LLMs always get this one wrong I have noticed so it's my go to, so maybe GPT 3.4? Or sub GPT-4 or on par if you don't care about translating things but I would say it's a stretch still. It did get some songs wrong at the end when I let it generate more, but it happened past a certain number and I noticed my context was set to 512 tokens so I think that was actually the issue not the model

from wizardlm.

I think this should be closed though since we know what the cause was now

from wizardlm.