[Bug] TVM failes to compile onnx model: TVMError: CodeGenVM cannot handle this intrinsic now: Op(relax.concat)
- Dominant language
- Python
- Stars
- 13.7k
- Forks
- 4k
- Avg merge
- 2d 19h
- Merged PRs (30d)
- 111
Description
### Expected behavior
TVM should build the model correctly.
### Actual behavior
When compiling the model, TVM crashes as follows:

```c
Traceback (most recent call last):
File "/home/carla/Documents/test/test.py", line 189, in
main()
File "/home/carla/Documents/test/test.py", line 178, in main
check_correctness(onnx_model)
File "/home/carla/Documents/test/test.py", line 126, in check_correctness
ex = relax.build(tvm_model, target="llvm", relax_pipeline=relax_pipeline)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/carla/Documents/tvm/python/tvm/relax/vm_build.py", line 258, in build
mod = _vmcodegen(builder, mod, exec_mode)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/carla/Documents/tvm/python/tvm/relax/vm_build.py", line 77, in _vmcodegen
return _ffi_api.VMCodeGen(builder, mod) # type:ignore
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "tvm/_ffi/_cython/./packed_func.pxi", line 339, in tvm._ffi._cy3.core.PackedFuncBase.__call__
File "tvm/_ffi/_cython/./packed_func.pxi", line 270, in tvm._ffi._cy3.core.FuncCall
File "tvm/_ffi/_cython/./packed_func.pxi", line 259, in tvm._ffi._cy3.core.FuncCall3
File "tvm/_ffi/_cython/./base.pxi", line 185, in tvm._ffi._cy3.core.CHECK_CALL
File "/home/carla/Documents/tvm/python/tvm/_ffi/base.py", line 468, in raise_last_ffi_error
raise py_err
tvm._ffi.base.TVMError: Traceback (most recent call last):
7: tvm::runtime::PackedFuncObj::Extractor::AssignTypedLambda(tvm::IRModule (*)(tvm::relax::ExecBuilder, tvm::IRModule), std::__cxx11::basic_string, std::allocator >)::{lambda(tvm::runtime::TVMArgs const&, tvm::runtime::TVMRetValue*)#1}> >::Call(tvm::runtime::PackedFuncObj const*, tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)
6: tvm::relax::relax_vm::VMCodeGen(tvm::relax::ExecBuilder, tvm::IRModule)
5: tvm::relax::relax_vm::CodeGenVM::Run(tvm::relax::ExecBuilder, tvm::IRModule)
4: tvm::relax::relax_vm::CodeGenVM::Codegen(tvm::relax::Function const&)
3: tvm::relax::ExprFunctor::VisitExpr(tvm::RelaxExpr const&)
2: tvm::relax::relax_vm::CodeGenVM::VisitExpr_(tvm::relax::SeqExprNode const*)
1: tvm::relax::ExprFunctor::VisitExpr(tvm::RelaxExpr const&)
0: tvm::relax::relax_vm::CodeGenVM::VisitExpr_(tvm::relax::CallNode const*)
File "/home/carla/Documents/tvm/src/relax/backend/vm/codegen_vm.cc", line 156
TVMError: CodeGenVM cannot handle this intrinsic now:
Op(relax.concat)
```
### Environment
OS: Ubuntu 20.04
TVM: 0.21.dev0(c00f52a70)
### Steps to reproduce
This bug can be reproduced by the following code with the model in the attachment. As shown in the code, the model can be executed by onnxruntime, which indicates that this is a valid model. It is strange that CodeGenVM cannot handle the concat operator.
```python
import sys
import numpy as np
import onnx
import onnxruntime
import tvm
from tvm import relax
from tvm.relax.frontend.onnx import from_onnx
import pickle
def main():
onnx_model = onnx.load("a317.onnx")
with open("inputs.pkl", "rb") as fp:
inputs = pickle.load(fp)
try:
ort_session = onnxruntime.InferenceSession(
onnx_model.SerializeToString(), providers=["CPUExecutionProvider"]
)
ort_output = ort_session.run([], inputs)
except Exception as e:
print(e)
sys.exit(1)
tvm_model = from_onnx(onnx_model, keep_params_in_input=True)
tvm_model = relax.transform.DecomposeOpsForInference()(tvm_model)
tvm_model = relax.transform.LegalizeOps()(tvm_model)
tvm_model, params = relax.frontend.detach_params(tvm_model)
with tvm.transform.PassContext(opt_level=0):
ex = relax.build(tvm_model, target="llvm")
if __name__ == "__main__":
main()
```
[testcase.zip](https://github.com/user-attachments/files/19830818/testcase.zip)
### Triage
* needs-triage
Contributor guide
No contributing guide indexed for this repository
Research direction
Start with src/relax/backend/vm/codegen_vm.cc at line 156 and reproduce the failure using the attached testcase and the provided Python script. Trace how CodeGenVM visits relax.concat, then run the reproduction and relevant Relax tests to verify that compiling the ONNX model no longer raises the unsupported-intrinsic error.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- compilers, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100