[Transform] set wdt/odt as float32 for CollapseRepeated

src/finn/transformation/streamline/collapse_repeated.py

@@ -30,6 +30,7 @@ from onnx import helper as oh
 
 from finn.transformation import Transformation
 from finn.transformation.infer_shapes import InferShapes
+from finn.core.datatype import DataType
 
 
 class CollapseRepeatedOp(Transformation):
@@ -83,6 +84,9 @@ class CollapseRepeatedOp(Transformation):
                     graph.node.insert(node_ind, new_node)
                     # replace parameter value
                     model.set_initializer(new_node_param_name, new_param)
+                    # be conservative with param/output DataTypes
+                    model.set_tensor_datatype(new_node_param_name, DataType.FLOAT32)
+                    model.set_tensor_datatype(end_name, DataType.FLOAT32)
                     # remove old nodes
                     graph.node.remove(n)
                     graph.node.remove(consumer)