I’m trying to use JAX’s custom_vjp to define custom gradient computations for a function that takes a SymPy expression as an input. However, I’m encountering errors because JAX doesn’t support non-JAX types as inputs for functions that are transformed (e.g., with grad, jit, or custom_vjp). I am recently modifying code in ScQubits to add a new jax backend to improve the efficiency, and then I meet some problem with jax and sympy.

I’m trying to modify the scqubits Python package to use JAX’s custom_vjp for differentiable programming. While doing so, I encountered the following error:

I’m trying to modify the scqubits Python package to use JAX’s custom_vjp for differentiable programming. While doing so, I encountered the following error:

I’m trying to modify the scqubits Python package to use JAX’s custom_vjp for differentiable programming. While doing so, I encountered the following error:

I’m trying to modify the scqubits Python package to use JAX’s custom_vjp for differentiable programming. While doing so, I encountered the following error:

I’m trying to modify the scqubits Python package to use JAX’s custom_vjp for differentiable programming. While doing so, I encountered the following error:

I’m trying to modify the scqubits Python package to use JAX’s custom_vjp for differentiable programming. While doing so, I encountered the following error: