f0d767eeccb697243e7f5b6fdf5570871c535ddc,opennmt/runner.py,Runner,_make_estimator,#Runner#,122

Before Change

        session_config=self._session_config,
        tf_random_seed=self._seed)

    if "train" in self._config:
      if "save_summary_steps" in self._config["train"]:
        accum = self._config["params"].get("gradients_accum", 1)
        summary_steps = self._config["train"]["save_summary_steps"]
        run_config = run_config.replace(
            save_summary_steps=summary_steps,
            log_step_count_steps=accum * summary_steps)
      if "save_checkpoints_steps" in self._config["train"]:
        run_config = run_config.replace(
            save_checkpoints_secs=None,
            save_checkpoints_steps=self._config["train"]["save_checkpoints_steps"])
      if "keep_checkpoint_max" in self._config["train"]:
        run_config = run_config.replace(
            keep_checkpoint_max=self._config["train"]["keep_checkpoint_max"])

    devices = get_devices(num_devices=self._num_devices, session_config=self._session_config)
    return tf.estimator.Estimator(
        self._model.model_fn(
            eval_prediction_hooks_fn=self._make_eval_prediction_hooks_fn(),

After Change

        tf_random_seed=self._seed,
        save_summary_steps=summary_steps,
        session_config=self._session_config,
        log_step_count_steps=params.get("gradients_accum", 1) * summary_steps)
    if "save_checkpoints_steps" in train_config:
      run_config = run_config.replace(
          save_checkpoints_secs=None,
          save_checkpoints_steps=train_config["save_checkpoints_steps"])
    if "keep_checkpoint_max" in train_config:
      run_config = run_config.replace(
          keep_checkpoint_max=train_config["keep_checkpoint_max"])

    devices = get_devices(num_devices=self._num_devices, session_config=self._session_config)
    return tf.estimator.Estimator(
        self._model.model_fn(

Instances