Merge branch 'main' into nested_model

Author: jmitrevs

.pre-commit-config.yaml

@@ -2,7 +2,7 @@ exclude: (^hls4ml\/templates\/(vivado|quartus)\/(ap_types|ac_types)\/|^test/pyte
 
 repos:
 - repo: https://github.com/psf/black
-  rev: 22.12.0
+  rev: 23.3.0
   hooks:
   - id: black
     language_version: python3
@@ -30,18 +30,18 @@ repos:
     args: ["--profile", "black", --line-length=125]
 
 - repo: https://github.com/asottile/pyupgrade
-  rev: v3.3.1
+  rev: v3.6.0
   hooks:
   - id: pyupgrade
     args: ["--py36-plus"]
 
 - repo: https://github.com/asottile/setup-cfg-fmt
-  rev: v2.2.0
+  rev: v2.3.0
   hooks:
   - id: setup-cfg-fmt
 
 - repo: https://github.com/pycqa/flake8
-  rev: 5.0.4
+  rev: 6.0.0
   hooks:
   - id: flake8
     exclude: docs/conf.py

contrib/kl_layer/kl_layer.py

@@ -34,7 +34,6 @@ def build(self, input_shape):
         super().build(input_shape)
 
     def _merge_function(self, inputs):
-
         mean = inputs[0]
         log_var = inputs[1]
 

hls4ml/backends/fpga/fpga_backend.py

@@ -391,7 +391,6 @@ def product_type(self, data_T, weight_T):
         return product
 
     def compute_conv1d_instructions(self, in_W, in_C, kernel_size=3, stride=1, pad=0):
-
         # Current limitations
         assert pad == 0
 
@@ -427,7 +426,6 @@ def compute_conv1d_instructions(self, in_W, in_C, kernel_size=3, stride=1, pad=0
         return (min_W, windows_int)
 
     def compute_conv2d_instructions(self, in_H, in_W, in_C, kernel_size=3, stride=1, pad=0):
-
         if isinstance(kernel_size, Iterable):
             kernel_height = kernel_size[0]
             kernel_width = kernel_size[1]

hls4ml/backends/quartus/passes/convolution_winograd.py

@@ -82,7 +82,6 @@ def match(self, node):
     def transform(self, model, node):
         if isinstance(node, Conv1D):
             if node.get_attr('filt_width', 3) == 3:
-
                 # First, transpose to a format suitable for the Winograd algorithm (F, C, W)
                 # Note, this assumes a format post-resource strategy optimizer, that is (F, W, C)
                 # Therefore, (F, W, C) => (F, C, W)
@@ -127,7 +126,6 @@ def transform(self, model, node):
 
         elif isinstance(node, Conv2D):
             if node.get_attr('filt_height', 3) == 3 and node.get_attr('filt_width', 3) == 3:
-
                 # First, transpose to a format suitable for the Winograd algorithm (F, C, H, W)
                 # Note, this assumes a format post-resource strategy optimizer, that is (F, H, W, C)
                 # Therefore, (F, H, W, C) => (F, C, H, W)

hls4ml/backends/quartus/quartus_backend.py

@@ -133,7 +133,6 @@ def create_initial_config(self, part='Arria10', clock_period=5, io_type='io_para
         return config
 
     def build(self, model, synth=True, fpgasynth=False, log_level=1, cont_if_large_area=False):
-
         """
         Builds the project using Intel HLS compiler.
 

hls4ml/backends/vivado/passes/convolution_templates.py

@@ -9,7 +9,7 @@
     static const unsigned n_out = {n_out};
     static const unsigned reuse_factor = {reuse};
     static const unsigned strategy = nnet::{strategy};
-    static const unsigned n_zeros = 0;
+    static const unsigned n_zeros = {nzeros};
     static const unsigned multiplier_limit = DIV_ROUNDUP(n_in * n_out, reuse_factor) - n_zeros / reuse_factor;
     typedef {accum_t.name} accum_t;
     typedef {bias_t.name} bias_t;
@@ -83,6 +83,7 @@ def format(self, node):
         mult_params = self._default_config_params(node)
         mult_params['n_in'] = node.get_attr('n_chan') * node.get_attr('filt_width')
         mult_params['n_out'] = node.get_attr('n_filt')
+        mult_params['nzeros'] = node.get_weights('weight').nzeros
         mult_params['product_type'] = get_backend('vivado').product_type(
             node.get_input_variable().type.precision, node.get_weights('weight').type.precision
         )
@@ -189,6 +190,7 @@ def format(self, node):
         mult_params = self._default_config_params(node)
         mult_params['n_in'] = node.get_attr('n_chan') * node.get_attr('filt_height') * node.get_attr('filt_width')
         mult_params['n_out'] = node.get_attr('n_filt')
+        mult_params['nzeros'] = node.get_weights('weight').nzeros
         mult_params['product_type'] = get_backend('vivado').product_type(
             node.get_input_variable().type.precision, node.get_weights('weight').type.precision
         )
@@ -274,6 +276,7 @@ def format(self, node):
         mult_params['index'] = str(node.index) + '_depthwise'
         mult_params['n_in'] = node.get_attr('n_chan') * node.get_attr('filt_width')
         mult_params['n_out'] = node.get_attr('n_chan')
+        mult_params['nzeros'] = node.get_weights('depthwise').nzeros
         mult_params['weight_t'] = node.get_weights('depthwise').type
         mult_params['product_type'] = get_backend('vivado').product_type(
             node.get_input_variable().type.precision, node.get_weights('depthwise').type.precision
@@ -313,6 +316,7 @@ def format(self, node):
         mult_params['index'] = str(node.index) + '_pointwise'
         mult_params['n_in'] = node.get_attr('n_chan')
         mult_params['n_out'] = node.get_attr('n_filt')
+        mult_params['nzeros'] = node.get_weights('pointwise').nzeros
         mult_params['weight_t'] = node.get_weights('pointwise').type
         mult_params['product_type'] = get_backend('vivado').product_type(
             node.get_input_variable().type.precision, node.get_weights('pointwise').type.precision
@@ -395,6 +399,7 @@ def format(self, node):
         mult_params['index'] = str(node.index) + '_depthwise'
         mult_params['n_in'] = node.get_attr('n_chan') * node.get_attr('filt_height') * node.get_attr('filt_width')
         mult_params['n_out'] = node.get_attr('n_chan')
+        mult_params['nzeros'] = node.get_weights('depthwise').nzeros
         mult_params['weight_t'] = node.get_weights('depthwise').type
         mult_params['product_type'] = get_backend('vivado').product_type(
             node.get_input_variable().type.precision, node.get_weights('depthwise').type.precision
@@ -438,6 +443,7 @@ def format(self, node):
         mult_params['index'] = str(node.index) + '_pointwise'
         mult_params['n_in'] = node.get_attr('n_chan')
         mult_params['n_out'] = node.get_attr('n_filt')
+        mult_params['nzeros'] = node.get_weights('pointwise').nzeros
         mult_params['weight_t'] = node.get_weights('pointwise').type
         mult_params['product_type'] = get_backend('vivado').product_type(
             node.get_input_variable().type.precision, node.get_weights('pointwise').type.precision

hls4ml/backends/vivado/passes/recurrent_templates.py

@@ -79,7 +79,6 @@ def __init__(self):
         self.mult2_template = recr_mult_config_template
 
     def format(self, node):
-
         params = self._default_config_params(node)
 
         params['n_in'] = node.get_input_variable().dim_names[1]

hls4ml/backends/vivado/passes/resource_strategy.py

@@ -8,7 +8,6 @@ class ApplyResourceStrategy(OptimizerPass):
     '''Transposes the weights to use the dense_resource matrix multiply routine'''
 
     def match(self, node):
-
         node_matches = isinstance(node, (Dense, Conv1D, SeparableConv1D, Conv2D, SeparableConv2D, LSTM, GRU))
         is_resource_strategy = node.get_attr('strategy', '').lower() == 'resource'
         already_transformed = node.get_attr('_weights_transposed', False) is True

hls4ml/converters/__init__.py

@@ -34,14 +34,6 @@
     warnings.warn("WARNING: ONNX converter is not enabled!", stacklevel=1)
     __onnx_enabled__ = False
 
-try:
-    from hls4ml.converters.tf_to_hls import tf_to_hls
-
-    __tensorflow_enabled__ = True
-except ImportError:
-    warnings.warn("WARNING: Tensorflow converter is not enabled!", stacklevel=1)
-    __tensorflow_enabled__ = False
-
 # ----------Layer handling register----------#
 model_types = ['keras', 'pytorch', 'onnx']
 
@@ -57,7 +49,6 @@
                 # and is defined in this module (i.e., not imported)
                 if callable(func) and hasattr(func, 'handles') and func.__module__ == lib.__name__:
                     for layer in func.handles:
-
                         if model_type == 'keras':
                             register_keras_layer_handler(layer, func)
                         elif model_type == 'pytorch':
@@ -139,11 +130,6 @@ def convert_from_config(config):
             model = pytorch_to_hls(yamlConfig)
         else:
             raise Exception("PyTorch not found. Please install PyTorch.")
-    elif 'TensorFlowModel' in yamlConfig:
-        if __tensorflow_enabled__:
-            model = tf_to_hls(yamlConfig)
-        else:
-            raise Exception("TensorFlow not found. Please install TensorFlow.")
     else:
         model = keras_to_hls(yamlConfig)
 

hls4ml/converters/keras/convolution.py

@@ -44,7 +44,7 @@ def parse_conv2d_layer(keras_layer, input_names, input_shapes, data_reader):
 
     (layer['in_height'], layer['in_width'], layer['n_chan']) = parse_data_format(input_shapes[0], layer['data_format'])
 
-    if layer['class_name'] in ['Conv2D', 'QConv2D']:
+    if layer['class_name'] in ['Conv2D', 'QConv2D', 'QConv2DBatchnorm']:
         layer['weight_data'] = get_weights_data(data_reader, layer['name'], 'kernel')
     elif layer['class_name'] in ['SeparableConv2D', 'QSeparableConv2D']:
         layer['depthwise_data'], layer['pointwise_data'] = get_weights_data(