hls4ml
hls4ml copied to clipboard
fastmachinelearning
Why am I getting this error : ´AttributeError: 'NoneType' object has no attribute 'shape'´ ?
Hello, I trained a ResNet model on Keras, then I followed the getting started a notebook to use hls4ml for research purposes. First, please check the config output using the following code:
config = hls4ml.utils.config_from_keras_model(model, granularity='model')
print_dict(config)
Model
Precision: ap_fixed<16,6>
ReuseFactor: 1
Strategy: Latency
-----------------------------------
Interpreting Model
Topology:
Layer name: input_2, layer type: InputLayer, input shapes: [[None, 1024, 2]], output shape: [None, 1024, 2]
Layer name: reshape_1, layer type: Reshape, input shapes: [[None, 1024, 2]], output shape: [None, 1, 1024, 2]
Layer name: ReStk1_conv1, layer type: Conv2D, input shapes: [[None, 1, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: ReStk1_conv2, layer type: Conv2D, input shapes: [[None, 32, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: ReStk1_conv3, layer type: Conv2D, input shapes: [[None, 32, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: add_12, layer type: Merge, input shapes: [[None, 32, 1024, 2], [None, 32, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: activation_13, layer type: Activation, input shapes: [[None, 32, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: ReStk1_conv4, layer type: Conv2D, input shapes: [[None, 32, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: ReStk1_conv5, layer type: Conv2D, input shapes: [[None, 32, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: add_13, layer type: Merge, input shapes: [[None, 32, 1024, 2], [None, 32, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: activation_14, layer type: Activation, input shapes: [[None, 32, 1024, 2]], output shape: [None, 32, 1024, 2]
Layer name: max_pooling2d_6, layer type: MaxPooling2D, input shapes: [[None, 32, 1024, 2]], output shape: [None, 32, 512, 1]
Layer name: ReStk2_conv1, layer type: Conv2D, input shapes: [[None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: ReStk2_conv2, layer type: Conv2D, input shapes: [[None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: ReStk2_conv3, layer type: Conv2D, input shapes: [[None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: add_14, layer type: Merge, input shapes: [[None, 32, 512, 1], [None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: activation_15, layer type: Activation, input shapes: [[None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: ReStk2_conv4, layer type: Conv2D, input shapes: [[None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: ReStk2_conv5, layer type: Conv2D, input shapes: [[None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: add_15, layer type: Merge, input shapes: [[None, 32, 512, 1], [None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: activation_16, layer type: Activation, input shapes: [[None, 32, 512, 1]], output shape: [None, 32, 512, 1]
Layer name: max_pooling2d_7, layer type: MaxPooling2D, input shapes: [[None, 32, 512, 1]], output shape: [None, 32, 256, 1]
Layer name: ReStk3_conv1, layer type: Conv2D, input shapes: [[None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: ReStk3_conv2, layer type: Conv2D, input shapes: [[None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: ReStk3_conv3, layer type: Conv2D, input shapes: [[None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: add_16, layer type: Merge, input shapes: [[None, 32, 256, 1], [None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: activation_17, layer type: Activation, input shapes: [[None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: ReStk3_conv4, layer type: Conv2D, input shapes: [[None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: ReStk3_conv5, layer type: Conv2D, input shapes: [[None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: add_17, layer type: Merge, input shapes: [[None, 32, 256, 1], [None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: activation_18, layer type: Activation, input shapes: [[None, 32, 256, 1]], output shape: [None, 32, 256, 1]
Layer name: max_pooling2d_8, layer type: MaxPooling2D, input shapes: [[None, 32, 256, 1]], output shape: [None, 32, 128, 1]
Layer name: ReStk4_conv1, layer type: Conv2D, input shapes: [[None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: ReStk4_conv2, layer type: Conv2D, input shapes: [[None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: ReStk4_conv3, layer type: Conv2D, input shapes: [[None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: add_18, layer type: Merge, input shapes: [[None, 32, 128, 1], [None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: activation_19, layer type: Activation, input shapes: [[None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: ReStk4_conv4, layer type: Conv2D, input shapes: [[None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: ReStk4_conv5, layer type: Conv2D, input shapes: [[None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: add_19, layer type: Merge, input shapes: [[None, 32, 128, 1], [None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: activation_20, layer type: Activation, input shapes: [[None, 32, 128, 1]], output shape: [None, 32, 128, 1]
Layer name: max_pooling2d_9, layer type: MaxPooling2D, input shapes: [[None, 32, 128, 1]], output shape: [None, 32, 64, 1]
Layer name: ReStk5_conv1, layer type: Conv2D, input shapes: [[None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: ReStk5_conv2, layer type: Conv2D, input shapes: [[None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: ReStk5_conv3, layer type: Conv2D, input shapes: [[None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: add_20, layer type: Merge, input shapes: [[None, 32, 64, 1], [None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: activation_21, layer type: Activation, input shapes: [[None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: ReStk5_conv4, layer type: Conv2D, input shapes: [[None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: ReStk5_conv5, layer type: Conv2D, input shapes: [[None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: add_21, layer type: Merge, input shapes: [[None, 32, 64, 1], [None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: activation_22, layer type: Activation, input shapes: [[None, 32, 64, 1]], output shape: [None, 32, 64, 1]
Layer name: max_pooling2d_10, layer type: MaxPooling2D, input shapes: [[None, 32, 64, 1]], output shape: [None, 32, 32, 1]
Layer name: ReStk6_conv1, layer type: Conv2D, input shapes: [[None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: ReStk6_conv2, layer type: Conv2D, input shapes: [[None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: ReStk6_conv3, layer type: Conv2D, input shapes: [[None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: add_22, layer type: Merge, input shapes: [[None, 32, 32, 1], [None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: activation_23, layer type: Activation, input shapes: [[None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: ReStk6_conv4, layer type: Conv2D, input shapes: [[None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: ReStk6_conv5, layer type: Conv2D, input shapes: [[None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: add_23, layer type: Merge, input shapes: [[None, 32, 32, 1], [None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: activation_24, layer type: Activation, input shapes: [[None, 32, 32, 1]], output shape: [None, 32, 32, 1]
Layer name: max_pooling2d_11, layer type: MaxPooling2D, input shapes: [[None, 32, 32, 1]], output shape: [None, 32, 16, 1]
Layer name: flatten_1, layer type: Reshape, input shapes: [[None, 32, 16, 1]], output shape: [None, 512]
Layer name: dense1, layer type: Dense, input shapes: [[None, 512]], output shape: [None, 128]
Layer name: dense2, layer type: Dense, input shapes: [[None, 128]], output shape: [None, 128]
Layer name: dense3, layer type: Dense, input shapes: [[None, 128]], output shape: [None, 7]
Layer name: activation_25, layer type: Softmax, input shapes: [[None, 7]], output shape: [None, 7]
However, When I use the convert_from_keras_model() as follows:
hls_model = hls4ml.converters.convert_from_keras_model(model,
hls_config=config,
output_dir='model_1/hls4ml_prj',
part='xczu7ev-ffvc1156-2-e')
I get the following errors:
Creating HLS model
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
/tmp/ipykernel_764/2595051309.py in <module>
17 print_dict(config)
18 print("-----------------------------------")
---> 19 hls_model = hls4ml.converters.convert_from_keras_model(model,
20 hls_config=config,
21 output_dir='model_1/hls4ml_prj',
/usr/local/lib/python3.8/dist-packages/hls4ml/converters/__init__.py in convert_from_keras_model(model, output_dir, project_name, input_data_tb, output_data_tb, backend, board, part, clock_period, io_type, hls_config, **kwargs)
218 _check_hls_config(config, hls_config)
219
--> 220 return keras_to_hls(config)
221
222
/usr/local/lib/python3.8/dist-packages/hls4ml/converters/keras_to_hls.py in keras_to_hls(config)
337
338 print('Creating HLS model')
--> 339 hls_model = HLSModel(config, reader, layer_list, input_layers, output_layers)
340 return hls_model
/usr/local/lib/python3.8/dist-packages/hls4ml/model/hls_model.py in __init__(self, config, data_reader, layer_list, inputs, outputs)
316 self._make_graph(layer_list)
317
--> 318 self._optimize_model(self.config.optimizers)
319
320 def _make_graph(self, layer_list):
/usr/local/lib/python3.8/dist-packages/hls4ml/model/hls_model.py in _optimize_model(self, optimizers)
334
335 def _optimize_model(self, optimizers):
--> 336 optimize_model(self, optimizers)
337
338 def make_node(self, kind, name, attributes, inputs, outputs=None):
/usr/local/lib/python3.8/dist-packages/hls4ml/model/optimizer/optimizer.py in optimize_model(model, passes)
36 for opt in optimizers:
37 for node in model.graph.values():
---> 38 if opt.match(node):
39 res = opt.transform(model, node)
40 if res:
/usr/local/lib/python3.8/dist-packages/hls4ml/model/optimizer/passes/repack_stream.py in match(self, node)
95 inp1 = node.get_input_variable(node.inputs[0])
96 inp2 = node.get_input_variable(node.inputs[1])
---> 97 return inp1.shape != inp2.shape
98 else:
99 return False
AttributeError: 'NoneType' object has no attribute 'shape'
Could someone explain to me what is the problem here and how to fix it? Thank you
Hello @AnouarITI
What version of hls4ml are you using?
Could you send a script or gist to reproduce the error?
Some recent updates may help this, e.g. https://github.com/fastmachinelearning/hls4ml/pull/443 https://github.com/fastmachinelearning/hls4ml/pull/472
Hello @jmduarte please find attached a notebook with the script to reproduce the error. Sorry, I can not share the dataset but at least it can give you more visibility of the error. script.txt
I have the same error when I use "add layers" in my model https://github.com/fastmachinelearning/hls4ml/issues/578
