Mirrors of deleted WizardLM models
ethanc8 opened this issue · comments
It seems that WizardLM has deleted all of their models. I'm posting here a few of the mirrors of these models. Feel free to add more mirrors as you find them -- you can copy the text of this post if you want.
I have not verified that the following mirrors are identical to the original models. Please use with caution.
If you find any more, please leave a comment here. If you have any on your hard disk and have sufficient bandwidth and time to upload them to HuggingFace, that would be greatly appreciated. Note that this focuses on bf16 versions of the models, in pytorch pickles or in safetensors, as those are closest to the original.
WizardLM-2
- WizardLM-2-7B, based on Mistral-7B-v0.1
- WizardLM-2-8x22B, based on Mixtral-8x22B
- WizardLM-2-70B, based on Llama2-70B, was never released
- GitHub repo
- Blog post
WizardCoder
- Paper on arXiv
- WizardCoder-33B-V1.1 (model card)
- WizardCoder-Python-34B-V1.0, based on CodeLlama-34b, does not have a mirror (model card)
- WizardCoder-15-V1.0, based on StarCoder-15B (fp16) (model card)
- WizardCoder-Python-13B-V1.0
- WizardCoder-Python-7B-V1.0
- WizardCoder-3B-V1.0
- WizardCoder-1B-V1.0
WizardLM
- Dataset
WizardLM_evol_instruct_V2_196k
- Dataset
WizardLM_evol_instruct_70k
All of the following ones are based on some version of Llama2. - WizardLM-70B-V1.0
- WizardLM-13B-V1.2
- WizardLM-13B-V1.1
- WizardLM-30B-V1.0
- WizardLM-13B-V1.0
- WizardLM-7B-V1.0
WizardMath
- WizardMath-7B-V1.1
- WizardMath-70B-V1.0
- WizardMath-13B-V1.0
- WizardMath-7B-V1.0
Also, given Microsoft's recent rash of deletions, it might be good to archive other Microsoft models such as WaveCoder and Phi-2.