mct_quantizers
mct_quantizers copied to clipboard
TF 2.16 support
In TF 2.16 the add_weight
function is change
https://github.com/sony/mct_quantizers/blob/4bbc18e843ff04d1f46a2a16d5bace32b7b1b37c/mct_quantizers/keras/quantize_wrapper.py#L310
logs
self.optimizer_step = self.add_weight(
^^^^^^^^^^^^^^^^
ValueError: Cannot convert '('o', 'p', 't', 'i', 'm', 'i', 'z', 'e', 'r', '_', 's', 't', 'e', 'p')' to a shape. Found invalid entry 'o' of type '<class 'str'>'.
debug code
import inspect
signature = inspect.signature(self.add_weight)
print(signature)
(shape=None, initializer=None, dtype=None, trainable=True, regularizer=None, constraint=None, name=None)