fix proxy conversionn when there are garbages connected

calad0i · Nov 30, 2023 · 72d52fd · 72d52fd
1 parent 8e95f8b
commit 72d52fd
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 1 deletion.
diff --git a/src/HGQ/proxy/convert.py b/src/HGQ/proxy/convert.py
@@ -41,6 +41,8 @@ def solve_dependencies(model: keras.Model):
         if node.is_input:
             continue
         layer = node.layer
+        if layer not in model.layers:
+            continue
         requires = list(node.parent_nodes)
         provides = node
         dependencies_list.append((layer, requires, provides))

diff --git a/test/test_syn_hlayers.py b/test/test_syn_hlayers.py
@@ -26,7 +26,13 @@ def create_model(layer: str, rnd_strategy: str, io_type: str):
         _inp = HQuantize()(inp)
     else:
         raise Exception(f'Please add test for {layer}')
-    out = eval('HGQ.layers.' + layer)(_inp)
+
+    _layer = eval('HGQ.layers.' + layer)
+    if hasattr(_layer, 'bias') and _layer.bias is not None:
+        bias: tf.Variable = _layer.bias
+        bias.assign(tf.constant(np.random.uniform(-4, 4, _layer.bias.shape).astype(np.float32)))
+
+    out = _layer(_inp)
     model = keras.Model(inp, out)
 
     for layer in model.layers:

diff --git a/test/test_syn_large.py b/test/test_syn_large.py
@@ -52,6 +52,9 @@ def create_model(rnd_strategy: str, io_type: str):
         if hasattr(layer, 'paq'):
             fbw: tf.Variable = layer.paq.fbw
             fbw.assign(tf.constant(np.random.uniform(2, 6, fbw.shape).astype(np.float32)))
+        if hasattr(layer, 'bias') and layer.bias is not None:
+            bias: tf.Variable = layer.bias
+            bias.assign(tf.constant(np.random.uniform(-4, 4, layer.bias.shape).astype(np.float32)))
     return horrible_model
 
 

diff --git a/test/test_syn_small.py b/test/test_syn_small.py
@@ -39,6 +39,9 @@ def create_model(rnd_strategy: str, io_type: str):
         if hasattr(layer, 'paq'):
             fbw: tf.Variable = layer.paq.fbw
             fbw.assign(tf.constant(np.random.uniform(2, 6, fbw.shape).astype(np.float32)))
+        if hasattr(layer, 'bias') and layer.bias is not None:
+            bias: tf.Variable = layer.bias
+            bias.assign(tf.constant(np.random.uniform(-4, 4, layer.bias.shape).astype(np.float32)))
     return model