Do I need full TVM in order to USE_GRAPH_RUNTIME_DEBUG ON?


#1

I built tvm runtime with USE_GRAPH_RUNTIME_DEBUG ON

But model.run throws the following

Traceback (most recent call last):
  File "./test_sares_tvm.py", line 40, in <module>
    module.run(input=data)
  File "/usr/local/lib/python3.5/dist-packages/tvm-0.6.dev0-py3.5-linux-armv7l.egg/tvm/contrib/debugger/debug_runtime.py", line 246, in run
    self.debug_datum.dump_output_tensor()
  File "/usr/local/lib/python3.5/dist-packages/tvm-0.6.dev0-py3.5-linux-armv7l.egg/tvm/contrib/debugger/debug_result.py", line 153, in dump_output_tensor
    param_f.write(save_tensors(output_tensors))
  File "/usr/local/lib/python3.5/dist-packages/tvm-0.6.dev0-py3.5-linux-armv7l.egg/tvm/contrib/debugger/debug_result.py", line 260, in save_tensors
    _save_tensors = tvm.get_global_func("_save_param_dict")
  File "/usr/local/lib/python3.5/dist-packages/tvm-0.6.dev0-py3.5-linux-armv7l.egg/tvm/_ffi/function.py", line 239, in get_global_func
    raise ValueError("Cannot find global function %s" % name)
ValueError: Cannot find global function _save_param_dict

I found that

_save_param_dict = tvm.get_global_func("tvm.relay._save_param_dict")

So, _save_param_dict needs tvm.relay which is not included to tvm runtime

Do I need full tvm version in order to USE_GRAPH_RUNTIME_DEBUG ON on edge device?