TVMError: Check failed: NodeTypeChecker<TNodeRef>: :Check(sptr.get()): Expected type relay.Function but get relay.Module

Prologueyan · June 24, 2019, 8:11am

I have met this problem recently:
Traceback (most recent call last):
File “insightface_tvm.py”, line 38, in
relay.build(net, target, params=params)
File “/home/jnet/.local/lib/python3.6/site-packages/tvm-0.6.dev0-py3.6-linux-aarch64.egg/tvm/relay/build_module.py”, line 196, in build
params)
File “/home/jnet/.local/lib/python3.6/site-packages/tvm-0.6.dev0-py3.6-linux-aarch64.egg/tvm/relay/build_module.py”, line 107, in build
self._build(func, target, target_host)
File “tvm/_ffi/_cython/./function.pxi”, line 310, in tvm._ffi._cy2.core.FunctionBase.call
File “tvm/_ffi/_cython/./function.pxi”, line 245, in tvm._ffi._cy2.core.FuncCall
File “tvm/_ffi/_cython/./function.pxi”, line 234, in tvm._ffi._cy2.core.FuncCall3
File “tvm/_ffi/_cython/./base.pxi”, line 170, in tvm._ffi._cy2.core.CALL
tvm._ffi.base.TVMError: Traceback (most recent call last):
[bt] (4) /home/jnet/.local/lib/python3.6/site-packages/tvm-0.6.dev0-py3.6-linux-aarch64.egg/tvm/libtvm.so(TVMFuncCall+0x70) [0x7f779daf48]
[bt] (3) /home/jnet/.local/lib/python3.6/site-packages/tvm-0.6.dev0-py3.6-linux-aarch64.egg/tvm/libtvm.so(std::_Function_handler<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::relay::backend::RelayBuildModule::GetFunction(std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::shared_ptr<tvm::runtime:: ModuleNode> const&)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#3}>::_M_invoke(std::_Any_data const&, tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&)+0x2c) [0x7f77574744]
[bt] (2) /home/jnet/.local/lib/python3.6/site-packages/tvm-0.6.dev0-py3.6-linux-aarch64.egg/tvm/libtvm.so(tvm::relay::backend::RelayBuildModule::GetFunction(std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, std::shared_ptr<tvm::runtime: :ModuleNode> const&)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#3}::operator()(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const+0x1a4) [0x7f77574214]
[bt] (1) /home/jnet/.local/lib/python3.6/site-packages/tvm-0.6.dev0-py3.6-linux-aarch64.egg/tvm/libtvm.so(tvm::relay::Function tvm::runtime::TVMArgValue::AsNodeRef<tvm::relay:: Function>() const+0x26c) [0x7f7756c6f4]
[bt] (0) /home/jnet/.local/lib/python3.6/site-packages/tvm-0.6.dev0-py3.6-linux-aarch64.egg/tvm/libtvm.so(dmlc::LogMessageFatal::~LogMessageFatal()+0x4c) [0x7f772363a4]
File “/home/jnet/tvm/include/tvm/packed_func_ext.h”, line 141
TVMError: Check failed: NodeTypeChecker: :Check(sptr.get()): Expected type relay.Function but get relay.Module

Please help me!

justdolearning · June 25, 2019, 7:44am

I update the lastest code ,and encounter the same error when convert onnx to nnvm model. and when I rollback the code to old version , the error disappear .
such as:
git reset --hard 579e96da4481a

Prologueyan · June 25, 2019, 7:57am

Thanks for your reply！
I will try it

kevinthesun · June 25, 2019, 11:42pm

github.com/apache/tvm

Add mod supoort in relay.build

master ← apivovarov:relay_build_mod

opened 11:05PM - 23 Jun 19 UTC

apivovarov

+33 -23

``` # in order to compile models we usually call func, params = relay.frontend….from_<framework>(model_artifact, shape, ...) # and then we call graph, lib, params = relay.build(func, target, params=params) ``` Recently it was a change which changes `from_<framework>` output to return `mod` instead of `func` #3353 But `relay.build` still supports `func` as an input. This is why all existing users' scripts to compile models should be updated. All people need to add extra line to their scripts before `relay.build` ``` func = mod[mod.entry_func] ``` In order to avoid all people fixing their compile scripts I think we need to add support for `mod` input to `relay.build`. This PR adds `mod` input support to `relay.build` and updates TFLite tutorial. I can update other tutorials if we agree with this change.