Does this text match what is used elsewhere? I would word it more like "if axes contains multiple entries referring to the same dimension..." or something like that, since a positive and negative axis can refer to the same dimension but not actually be a "repeated entry".

That's a good point. No, as I noted in #747 I think this is the very first time that we've ever cared about (effectively) repeating axes in the standard. I'll update.

The other functions with axes (as opposed to axis) are

• permute_dims: this doesn't allow negative axes, and already says it should be "a permutation" which implies no duplicates
• tensordot: Doesn't say anything. Do duplicate axes make sense in this function?

Thanks, @leofang! If the suggestions are good for you, this should be ready for merge after resolution.

Thanks, @kgryte, suggestions applied.