Traceback (most recent call last):
File “/opt/software/tvm/vta/tests/python/integration/resnet2.py”, line 219, in
device)
File “/opt/software/tvm/vta/tests/python/integration/resnet2.py”, line 95, in generate_graph
params = nnvm.compiler.load_param_dict(t1)
File “/opt/software/tvm/nnvm/python/nnvm/compiler/param_dict.py”, line 60, in load_param_dict
load_arr = _load_param_dict(param_bytes)
File “/opt/software/tvm/python/tvm/_ffi/_ctypes/function.py”, line 184, in call
ctypes.byref(ret_val), ctypes.byref(ret_tcode)))
File “/opt/software/tvm/python/tvm/_ffi/base.py”, line 66, in check_call
raise TVMError(py_str(_LIB.TVMGetLastError()))
tvm._ffi.base.TVMError: [10:34:57] /opt/software/tvm/tvm/nnvm/src/compiler/graph_runtime.cc:79: Check failed: header == kTVMNDArrayListMagic Invalid parameters file format
Currently parameter files have to go through graph conversion to a format that is compatible with VTA (8-bit weight, packing etc.). This conversion was achieved with an internal branch which we decided not to support since Relay (the new graph IR) is underway and will provide much needed enhancements over NNVM.
As a result we are working on graph support for VTA so users can offload the models of their choosing. Please stay tuned for the VTA update that includes Relay support so you can run your SSD Resnet-50 model.
See the following issue: https://github.com/dmlc/tvm/issues/1673