Hi I saw the following table under the documentation for the docker toolchain. My tf/keras model includes a multipication layer, but the Table does not explicitly mention if it is compatible with KL520. Is there a work around to multiply two tensors in KL520?
Comments
Hi Anoop,
Kneron's hardware architecture mainly depends on the onnx specification.
You mentioned that your Keras model has multiply layer, which is the Mul operator in onnx.
Preliminary judgment, your multiply layer is not supported in KL520, but it is supported in KL720.
You can refer to the following two links:
(2.3 Supported operators) http://doc.kneron.com/docs/#toolchain/manual/
(Table 1.2 Operators conversion table for every platform) http://doc.kneron.com/docs/#toolchain/converters/