Description
padding='same' is supported by PyTorch but .....
With too many bugs, I decided to give up [intel-extension-for-pytorch].
RuntimeError Traceback (most recent call last)
Cell In[32], line 1
----> 1 models = train_models()
Cell In[31], line 27, in train_models()
25 model = ResNet50()
26 # model = ipex.optimize(model, dtype=torch.float32)
---> 27 trained_models = train_hyperparams(model)
29 models.extend(trained_models)
31 writer.close()
Cell In[31], line 14, in train_hyperparams(model)
12 # optimizer = torch.optim.Adam(model.parameters(), lr=lr, eps=1e-07)
13 optimizer = torch.optim.SGD(model.parameters(), lr=lr, momentum=momentum)
---> 14 model, optimizer = ipex.optimize(model, optimizer=optimizer)
15 loss_fn = nn.CrossEntropyLoss()
16 model, train_name, seconds = train_loop(train_loader, model, optimizer,loss_fn, loop)
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/frontend.py:294, in optimize(model, dtype, optimizer, level, inplace, conv_bn_folding, weights_prepack, replace_dropout_with_identity, optimize_lstm, split_master_weight_for_bf16, fuse_update_step, auto_kernel_selection, sample_input)
290 if dtype == torch.bfloat16:
291 assert core.onednn_has_bf16_support(),
292 "BF16 weight prepack needs the cpu support avx512bw, avx512vl and avx512dq, " +
293 "please set dtype to torch.float or set weights_prepack to False."
--> 294 optimized_model, optimized_optimizer, params_attr = utils._weight_prepack.weight_prepack_with_ipex(
295 optimized_model, optimized_optimizer, params_attr, opt_properties.auto_kernel_selection)
296 # TODO: model list, optimizer list.
297 if optimizer is None:
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/nn/utils/_weight_prepack.py:331, in weight_prepack_with_ipex(module, optimizer, params_attr, auto_kernel_selection)
328 setattr(new_m, name, convert_rec(sub_m, optimizer, params_attr, auto_kernel_selection)[0])
329 return new_m, optimizer, params_attr
--> 331 opt_model, opt_optmizer, params_attr = convert_rec(module, optimizer, params_attr, auto_kernel_selection)
332 if opt_optmizer is not None:
333 setattr(opt_optmizer, 'params_attr', params_attr)
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/nn/utils/_weight_prepack.py:328, in weight_prepack_with_ipex..convert_rec(m, optimizer, params_attr, auto_kernel_selection)
326 new_m = convert(m, optimizer, params_attr, auto_kernel_selection)
327 for name, sub_m in m.named_children():
--> 328 setattr(new_m, name, convert_rec(sub_m, optimizer, params_attr, auto_kernel_selection)[0])
329 return new_m, optimizer, params_attr
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/nn/utils/_weight_prepack.py:328, in weight_prepack_with_ipex..convert_rec(m, optimizer, params_attr, auto_kernel_selection)
326 new_m = convert(m, optimizer, params_attr, auto_kernel_selection)
327 for name, sub_m in m.named_children():
--> 328 setattr(new_m, name, convert_rec(sub_m, optimizer, params_attr, auto_kernel_selection)[0])
329 return new_m, optimizer, params_attr
[... skipping similar frames: weight_prepack_with_ipex.<locals>.convert_rec at line 328 (1 times)]
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/nn/utils/_weight_prepack.py:328, in weight_prepack_with_ipex..convert_rec(m, optimizer, params_attr, auto_kernel_selection)
326 new_m = convert(m, optimizer, params_attr, auto_kernel_selection)
327 for name, sub_m in m.named_children():
--> 328 setattr(new_m, name, convert_rec(sub_m, optimizer, params_attr, auto_kernel_selection)[0])
329 return new_m, optimizer, params_attr
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/nn/utils/_weight_prepack.py:326, in weight_prepack_with_ipex..convert_rec(m, optimizer, params_attr, auto_kernel_selection)
325 def convert_rec(m, optimizer, params_attr, auto_kernel_selection):
--> 326 new_m = convert(m, optimizer, params_attr, auto_kernel_selection)
327 for name, sub_m in m.named_children():
328 setattr(new_m, name, convert_rec(sub_m, optimizer, params_attr, auto_kernel_selection)[0])
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/nn/utils/_weight_prepack.py:293, in weight_prepack_with_ipex..convert(m, optimizer, params_attr, auto_kernel_selection)
291 if weight not in params_attr:
292 params_attr[weight] = {}
--> 293 new_m = IPEX_WEIGHT_PREPACK_MODULEtype(m)
294 params_attr[weight].update({
295 'op': type(m),
296 'ctx': new_m.ctx})
297 if hasattr(new_m, "weight_channels_last"):
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/nn/utils/_weight_prepack.py:86, in _IPEXConv2d.init(self, dense_module)
85 def init(self, dense_module):
---> 86 super(_IPEXConv2d, self).init(dense_module)
File /usr/local/lib/python3.8/dist-packages/intel_extension_for_pytorch/nn/utils/_weight_prepack.py:37, in _IPEXConvNd.init(self, dense_module)
35 self.register_parameter('bias', None)
36 # create conv op context
---> 37 self.ctx = torch.ops.ipex_prepack.convolution_prepack(
38 dense_module.weight, self.bias, self.stride, self.padding,
39 self.dilation, self.groups,
40 self.weight_channels_last, self.prepack_input_shape
41 )
43 self.weight = nn.Parameter(self.ctx.get_weight(), requires_grad = dense_module.weight.requires_grad)
45 # pack master_weight or weight_trail if needed
File /usr/local/lib/python3.8/dist-packages/torch/_ops.py:143, in OpOverloadPacket.call(self, *args, **kwargs)
138 def call(self, *args, **kwargs):
139 # overloading call to ensure torch.ops.foo.bar()
140 # is still callable from JIT
141 # We save the function ptr as the op attribute on
142 # OpOverloadPacket to access it here.
--> 143 return self._op(*args, **kwargs or {})
RuntimeError: ipex_prepack::convolution_prepack() Expected a value of type 'List[int]' for argument 'padding' but instead found type 'str'.
Position: 3
Value: 'same'
Declaration: ipex_prepack::convolution_prepack(Tensor W, Tensor? B, int[] stride, int[] padding, int[] dilation, int groups, bool input_is_channels_last, int[] input_sizes) -> (torch.torch.classes.ipex_prepack.ConvolutionOpContext)
Cast error details: Unable to cast Python instance to C++ type (compile in debug mode for details)