Hello everyone,
I would like to convert FP32 weight into FP16 weight
and inference by FP16 mode without tensorRT
described my process in the following
1,make a weight traind FP32 mode
2,convert weight to FP16 from FP32
weight_32 = model.get_weights()
weight_16 = []
for x in range(len(weight_32)):
weight_16.append(weight_32[x].astype(np.float16))
model.set_weights(weight_16)
3,inference by FP16
keras.backend.set_floatx("float16")
But it doesn’t run and shown follow message
>TypeError: Input 'b' of 'MatMul' Op has type float16 that does not match type float32 of argument 'a'.
Can I run such this method?
thanks,
best regards