Spaces:
Running
on
CPU Upgrade
Wrong Running Config of WizardLM-13B and WizardLM-30B
The finished job "victor123/WizardLM-13B-1.0", the running "WizardLM/WizardLM-13B-1.0" and the pending "WizardLM/WizardLM-30B-V1.0" are all delta weights of LLaMA weight, but I do not know who submitted them with the config of "is_delta_weight==False", that would result in a bad performance of evaluation. Could you please stop pending/running/showing these wrong jobs?
We can use "TheBloke/wizardLM-13B-1.0-fp16" and "TheBloke/WizardLM-30B-fp16" to evaluate WizardLM.
And it still scored a 48.6 on TruthfulQA somehow. Flawless benchmark.
Hi
@WizardLM
We can remove the models that are linked to your account from the display - did you resubmit the models as you wanted?
Closing as this discussion has been inactive and I'm quite sure we solved it in another thread, feel free to reopen if needed!