[lipefree]

The signature for arange is defined with a default argument (nb::none()) for the argument step in the runtime binding (python/src/ops.cpp), but the explicit signature used to generate stubs uses step: None | int | float which causes type checkers like ty to raise the error No overload of function arange matches arguments in case we try to specify the arguments start and stop, omitting step.

import mlx.core as mx

# Error from `ty`: No overload of function `arange` matches arguments
x = mx.arange(start=1, stop=2)

# But this passes type checking in `ty`
x = mx.arange(start=1, stop=2, step=1)

We fix the stub definition by defining the function as

def arange(..., step : None | int | float = None, ...)

Validation by rebuilding the Python extension, regenerating stubs and running ty as a type checker on the modified definition.

• ☑️ I understand it is strictly prohibited to use AI to write PR description
• AI usage disclosure: co-developed with coding agent for finding the fix, building the projects and running the tests