View source on GitHub

Writes a TensorFlow graph to the summary interface.

The graph summary is, strictly speaking, not a summary. Conditions like tf.summary.should_record_summaries do not apply. Only a single graph can be associated with a particular run. If multiple graphs are written, then only the last one will be considered by TensorBoard.

When not using eager execution mode, the user should consider passing the graph parameter to tf.compat.v1.summary.initialize instead of calling this function. Otherwise special care needs to be taken when using the graph to record the graph.

param A tf.Tensor containing a serialized graph proto. When eager execution is enabled, this function will automatically coerce tf.Graph, tf.compat.v1.GraphDef, and string types.
step The global step variable. This doesn't have useful semantics for graph summaries, but is used anyway, due to the structure of event log files. This defaults to the global step.
name A name for the operation (optional).

The created tf.Operation or a tf.no_op if summary writing has not been enabled for this context.

TypeError If param isn't already a tf.Tensor in graph mode.