search

jeshraghian / snntorch

Deep and online learning with spiking neural networks in Python
https://snntorch.readthedocs.io/en/latest/
MIT License
1.35k stars 222 forks source link

SNNTorch example not working on IPU #236

Open Markus-Goetz opened 1 year ago

Markus-Goetz commented 1 year ago

Description

We are trying to make the IPU tutorial work. The source can be found here:

https://github.com/jeshraghian/snntorch/blob/master/examples/tutorial_ipu_1.py

Unfortunately, it is not working for us in any shape or form, as soon as we include a LIF neuron in our model. Without any snntorch neurons the tutorial works perfectly.

What I Did

  1. Set up a virtual environment
  2. Install various dependencies, including Poplar SDK 3.1.0 and poptorch 1.13
  3. Install snntorch-ipu by cloning from source https://github.com/vinniesun/snntorch-ipu
  4. Manually fix the includes of the custom_ops based on this issue https://github.com/jeshraghian/snntorch/issues/153 to have the additional include #include <popart/popx/opx.hpp>
  5. Copy-paste your example code from above
  6. Run python with it

This results in the following stacktrace:

[00:09:05.182] [poptorch::python] [critical] NotImplementedError: Could not run 'aten::_local_scalar_dense' with arguments from the 'Meta' backend. This could be because the operator doesn't exist for this backend, or was omitted during the selective/custom build process (if using custom build). If you are a Facebook employee using PyTorch on mobile, please visit https://fburl.com/ptmfixes for possible resolutions. 'aten::_local_scalar_dense' is only available for these backends: [CPU, IPU, BackendSelect, Python, FuncTorchDynamicLayerBackMode, Functionalize, Named, Conjugate, Negative, ZeroTensor, ADInplaceOrView, AutogradOther, AutogradCPU, AutogradCUDA, AutogradHIP, AutogradXLA, AutogradMPS, AutogradIPU, AutogradXPU, AutogradHPU, AutogradVE, AutogradLazy, AutogradMeta, AutogradPrivateUse1, AutogradPrivateUse2, AutogradPrivateUse3, AutogradNestedTensor, Tracer, AutocastCPU, AutocastCUDA, FuncTorchBatched, FuncTorchVmapMode, Batched, VmapMode, FuncTorchGradWrapper, PythonTLSSnapshot, FuncTorchDynamicLayerFrontMode, PythonDispatcher].

CPU: registered at aten/src/ATen/RegisterCPU.cpp:30798 [kernel]
IPU: registered at /opt/jenkins/workspace/poptorch/poptorch_ci_rhel_8/unprivileged/poptorch/poptorch/source/dispatch_tracer/RegisterOptionalAtenOps.cpp.inc:3 [kernel]
BackendSelect: fallthrough registered at ../aten/src/ATen/core/BackendSelectFallbackKernel.cpp:3 [backend fallback]
Python: registered at ../aten/src/ATen/core/PythonFallbackKernel.cpp:140 [backend fallback]
FuncTorchDynamicLayerBackMode: registered at ../aten/src/ATen/functorch/DynamicLayer.cpp:488 [backend fallback]
Functionalize: registered at ../aten/src/ATen/FunctionalizeFallbackKernel.cpp:291 [backend fallback]
Named: fallthrough registered at ../aten/src/ATen/core/NamedRegistrations.cpp:11 [kernel]
Conjugate: registered at ../aten/src/ATen/ConjugateFallback.cpp:18 [backend fallback]
Negative: registered at ../aten/src/ATen/native/NegateFallback.cpp:18 [backend fallback]
ZeroTensor: registered at ../aten/src/ATen/ZeroTensorFallback.cpp:86 [backend fallback]
ADInplaceOrView: fallthrough registered at ../aten/src/ATen/core/VariableFallbackKernel.cpp:64 [backend fallback]
AutogradOther: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradCPU: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradCUDA: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradHIP: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradXLA: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradMPS: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradIPU: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradXPU: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradHPU: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradVE: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradLazy: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradMeta: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradPrivateUse1: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradPrivateUse2: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradPrivateUse3: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradNestedTensor: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
Tracer: registered at ../torch/csrc/autograd/generated/TraceType_2.cpp:16890 [kernel]
AutocastCPU: fallthrough registered at ../aten/src/ATen/autocast_mode.cpp:482 [backend fallback]
AutocastCUDA: fallthrough registered at ../aten/src/ATen/autocast_mode.cpp:324 [backend fallback]
FuncTorchBatched: registered at ../aten/src/ATen/functorch/BatchRulesDynamic.cpp:64 [kernel]
FuncTorchVmapMode: fallthrough registered at ../aten/src/ATen/functorch/VmapModeRegistrations.cpp:28 [backend fallback]
Batched: registered at ../aten/src/ATen/BatchingRegistrations.cpp:1064 [backend fallback]
VmapMode: fallthrough registered at ../aten/src/ATen/VmapModeRegistrations.cpp:33 [backend fallback]
FuncTorchGradWrapper: registered at ../aten/src/ATen/functorch/TensorWrapper.cpp:189 [backend fallback]
PythonTLSSnapshot: registered at ../aten/src/ATen/core/PythonFallbackKernel.cpp:148 [backend fallback]
FuncTorchDynamicLayerFrontMode: registered at ../aten/src/ATen/functorch/DynamicLayer.cpp:484 [backend fallback]
PythonDispatcher: registered at ../aten/src/ATen/core/PythonFallbackKernel.cpp:144 [backend fallback]

Traceback (most recent call last):
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/PETNet/ipu_tutorial.py", line 125, in <module>
    output, loss = poptorch_model(data, labels)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/venv/petnet-ipu/lib64/python3.9/site-packages/poptorch/_poplar_executor.py", line 1151, in __call__
    self._compile(in_tensors)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/venv/petnet-ipu/lib64/python3.9/site-packages/poptorch/_impl.py", line 358, in wrapper
    return func(self, *args, **kwargs)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/venv/petnet-ipu/lib64/python3.9/site-packages/poptorch/_poplar_executor.py", line 911, in _compile
    self._executable = self._compileWithDispatch(in_tensors_trace_view)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/venv/petnet-ipu/lib64/python3.9/site-packages/poptorch/_impl.py", line 164, in wrapper
    return func(*args, **kwargs)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/venv/petnet-ipu/lib64/python3.9/site-packages/poptorch/_poplar_executor.py", line 787, in _compileWithDispatch
    result = self._model(*args, **kwargs)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/venv/petnet-ipu/lib64/python3.9/site-packages/torch/nn/modules/module.py", line 1190, in _call_impl
    return forward_call(*input, **kwargs)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/PETNet/ipu_tutorial.py", line 80, in forward
    out, mem1 = self.lif1(x, mem1)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/venv/petnet-ipu/lib64/python3.9/site-packages/torch/nn/modules/module.py", line 1208, in _call_impl
    result = forward_call(*input, **kwargs)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/snntorch-ipu/snntorch/_neurons/leaky.py", line 146, in forward
    mem = self.state_fn(input_, mem)
  File "/hkfs/work/workspace/scratch/ih5525-PETNet/snntorch-ipu/snntorch/_neurons/leaky.py", line 182, in _build_state_function
    if self.reset_mechanism_val == 0:  # reset by subtraction
NotImplementedError: Could not run 'aten::_local_scalar_dense' with arguments from the 'Meta' backend. This could be because the operator doesn't exist for this backend, or was omitted during the selective/custom build process (if using custom build). If you are a Facebook employee using PyTorch on mobile, please visit https://fburl.com/ptmfixes for possible resolutions. 'aten::_local_scalar_dense' is only available for these backends: [CPU, IPU, BackendSelect, Python, FuncTorchDynamicLayerBackMode, Functionalize, Named, Conjugate, Negative, ZeroTensor, ADInplaceOrView, AutogradOther, AutogradCPU, AutogradCUDA, AutogradHIP, AutogradXLA, AutogradMPS, AutogradIPU, AutogradXPU, AutogradHPU, AutogradVE, AutogradLazy, AutogradMeta, AutogradPrivateUse1, AutogradPrivateUse2, AutogradPrivateUse3, AutogradNestedTensor, Tracer, AutocastCPU, AutocastCUDA, FuncTorchBatched, FuncTorchVmapMode, Batched, VmapMode, FuncTorchGradWrapper, PythonTLSSnapshot, FuncTorchDynamicLayerFrontMode, PythonDispatcher].

CPU: registered at aten/src/ATen/RegisterCPU.cpp:30798 [kernel]
IPU: registered at /opt/jenkins/workspace/poptorch/poptorch_ci_rhel_8/unprivileged/poptorch/poptorch/source/dispatch_tracer/RegisterOptionalAtenOps.cpp.inc:3 [kernel]
BackendSelect: fallthrough registered at ../aten/src/ATen/core/BackendSelectFallbackKernel.cpp:3 [backend fallback]
Python: registered at ../aten/src/ATen/core/PythonFallbackKernel.cpp:140 [backend fallback]
FuncTorchDynamicLayerBackMode: registered at ../aten/src/ATen/functorch/DynamicLayer.cpp:488 [backend fallback]
Functionalize: registered at ../aten/src/ATen/FunctionalizeFallbackKernel.cpp:291 [backend fallback]
Named: fallthrough registered at ../aten/src/ATen/core/NamedRegistrations.cpp:11 [kernel]
Conjugate: registered at ../aten/src/ATen/ConjugateFallback.cpp:18 [backend fallback]
Negative: registered at ../aten/src/ATen/native/NegateFallback.cpp:18 [backend fallback]
ZeroTensor: registered at ../aten/src/ATen/ZeroTensorFallback.cpp:86 [backend fallback]
ADInplaceOrView: fallthrough registered at ../aten/src/ATen/core/VariableFallbackKernel.cpp:64 [backend fallback]
AutogradOther: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradCPU: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradCUDA: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradHIP: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradXLA: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradMPS: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradIPU: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradXPU: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradHPU: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradVE: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradLazy: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradMeta: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradPrivateUse1: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradPrivateUse2: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradPrivateUse3: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
AutogradNestedTensor: registered at ../torch/csrc/autograd/generated/VariableType_2.cpp:16899 [autograd kernel]
Tracer: registered at ../torch/csrc/autograd/generated/TraceType_2.cpp:16890 [kernel]
AutocastCPU: fallthrough registered at ../aten/src/ATen/autocast_mode.cpp:482 [backend fallback]
AutocastCUDA: fallthrough registered at ../aten/src/ATen/autocast_mode.cpp:324 [backend fallback]
FuncTorchBatched: registered at ../aten/src/ATen/functorch/BatchRulesDynamic.cpp:64 [kernel]
FuncTorchVmapMode: fallthrough registered at ../aten/src/ATen/functorch/VmapModeRegistrations.cpp:28 [backend fallback]
Batched: registered at ../aten/src/ATen/BatchingRegistrations.cpp:1064 [backend fallback]
VmapMode: fallthrough registered at ../aten/src/ATen/VmapModeRegistrations.cpp:33 [backend fallback]
FuncTorchGradWrapper: registered at ../aten/src/ATen/functorch/TensorWrapper.cpp:189 [backend fallback]
PythonTLSSnapshot: registered at ../aten/src/ATen/core/PythonFallbackKernel.cpp:148 [backend fallback]
FuncTorchDynamicLayerFrontMode: registered at ../aten/src/ATen/functorch/DynamicLayer.cpp:484 [backend fallback]
PythonDispatcher: registered at ../aten/src/ATen/core/PythonFallbackKernel.cpp:144 [backend fallback]
vinniesun commented 1 year ago

@Markus-Goetz Thanks for letting us know of this error. I'll take a look at it to see if it's another Popart SDK update that we need to update our codebase to.

ahenkes1 commented 1 year ago

@vinniesun , are there any updates on this problem?

vinniesun commented 1 year ago

still discussing with Graphcore to find the root cause of this bug. The last working version of snntorch-ipu poplar was 2.x.

ahenkes1 commented 1 year ago

@vinniesun , thanks for the update!