jax

ce782e61 - Fixed bug with silent TF using 64-bit operations in 32-bit mode. (#3471)

Fixed bug with silent TF using 64-bit operations in 32-bit mode. (#3471) We already cast the NumPy arrays to 32-bit, but this was not happening if the input was tf.Variable or tf.constant. In the process I have added some invariant checks, if core.skip_checks is False, which it always is in testing. Added also some sanity checking or arguments.