If you are lazy just Control F and search “The webpage contains tons of valuable info about their training strategies” and jump there
As known by almost everyone now, Wizard-LM-2 was taken down for some mysterious toxicity testing https://twitter.com/WizardLM_AI/status/1780101465950105775
This tweet is dated April 16, 2024. Today is 25 April. A toxicity test does not take 2 weeks. This is suspicious.
They didn’t only take down Wizard-LM-2 models, they also took down their entire HF repo including older Wizard LM coder, Wizard LM math and Wizard LM models, their papers, links to their website, github and twitter. https://huggingface.co/WizardLM
When we goto March 24 version of their repo we can find all their older stuff to confirm it was indeed there before. (pre April 16 unfortunately unavailable on archive.org or archive.ph) https://web.archive.org/web/20240324124154/https://huggingface.co/WizardLM
If we follow one of the urls to the paper, the paper is indeed up and available at https://huggingface.co/papers/2304.12244 and https://arxiv.org/abs/2304.12244 but keep in mind that once published on arxiv, a paper cannot be taken down, only marked as withdrawn which still allows access. E.g. once on arxiv, always on arxiv.
This gives us valuable data. Here are all the involved individuals with the Wizard-LM 1 paper.