I load the model from onnx and use debug_get_output(nodename, out) from debug_runtime trying to get the Intermediate results. But all the Intermediate results are wrong except the final results. Here is my part of codes(Not complete for pithy).
from tvm.contrib.debugger import debug_runtime as graph_runtime
m.run()
inA = np.array([])
outT = []
for nodename in nodenames:
outT.append(m.debug_get_output('fused_nn_relu_4_0', inA).asnumpy())
eg:The relu operation in my onnx model is max(x, 0) but there are many negative numbers in the debug_runtime version relu results. Another example is that the final dense result is the same as the final softmax. So much unbeliveable results just happended. It’s weird and I can’t figure it out through my tough trying. Crying for help~plz