7a81491366
Removes a bit of boilerplate from training loops (making a prefix from a directory). Also clarifies the recovery of checkpoint lists (like tf.train.Saver.recover_last_checkpoints, but automatic and more thorough). Adds a couple fields to the CheckpointState proto to support this. Should live in contrib until I make it work well with tf.keras.Model.save_weights. When used together, save_weights needs to number its checkpoints. (There's a TODO for this.) PiperOrigin-RevId: 208566198
23 lines
769 B
Protocol Buffer
23 lines
769 B
Protocol Buffer
syntax = "proto3";
|
|
|
|
package tensorflow;
|
|
option cc_enable_arenas = true;
|
|
|
|
// Protocol buffer representing the checkpoint state.
|
|
message CheckpointState {
|
|
// Path to the most-recent model checkpoint.
|
|
string model_checkpoint_path = 1;
|
|
|
|
// Paths to all not-yet-deleted model checkpoints, sorted from oldest to
|
|
// newest.
|
|
// Note that the value of model_checkpoint_path should be the last item in
|
|
// this list.
|
|
repeated string all_model_checkpoint_paths = 2;
|
|
// Unix timestamps corresponding to all_model_checkpoint_paths, indicating
|
|
// when each checkpoint was created.
|
|
repeated double all_model_checkpoint_timestamps = 3;
|
|
// Unix timestamp indicating the creation time for the last preserved
|
|
// checkpoint.
|
|
double last_preserved_timestamp = 4;
|
|
}
|