tensorflow/models · error · ValueError

Number of transformer layer must be equal or divisible.

Error message

Number of transformer layer must be equal or divisible.

What it means

Error "Number of transformer layer must be equal or divisible." thrown in tensorflow/models.

Source

Thrown at official/projects/edgetpu/nlp/mobilebert_edgetpu_trainer.py:136

    self.word_vocab_size = experiment_params.student_model.encoder.mobilebert.word_vocab_size
    self.distill_gt_ratio = experiment_params.end_to_end_distillation.distill_ground_truth_ratio
    self.teacher_transformer_layers = experiment_params.teacher_model.encoder.mobilebert.num_blocks
    self.student_transformer_layers = experiment_params.student_model.encoder.mobilebert.num_blocks
    self.exported_ckpt_path = export_ckpt_path
    self.current_step = orbit.utils.create_global_step()
    self.current_step.assign(0)

    # Stage is updated every time when the distillation is done for one
    # transformer layer. self.stage is updated at the train_loop_begin()
    # function. After the last stage is done, the self.mode is changed to
    # 'e2e'.
    self.stage = 0
    self.mode = DistillationMode.INIT

    # Number of transformer layers in teacher should be equal (or divisible)
    # by the number of transformer layers in student.
    if self.teacher_transformer_layers % self.student_transformer_layers != 0:
      raise ValueError(
          'Number of transformer layer must be equal or divisible.')
    self.ratio = (self.teacher_transformer_layers //
                  self.student_transformer_layers)

    # Create optimizers for different training stage.
    self.layer_wise_optimizer = self.build_optimizer(
        self.layer_wise_distill_config)
    self.e2e_optimizer = self.build_optimizer(self.e2e_distill_config)
    self.current_optimizer = self.layer_wise_optimizer

    # A non-trainable layer for feature normalization for transfer loss.
    self._layer_norm = tf_keras.layers.LayerNormalization(
        axis=-1,
        beta_initializer='zeros',
        gamma_initializer='ones',
        trainable=False)

    self.build_dataset()

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/projects/edgetpu/nlp/mobilebert_edgetpu_trainer.py:136 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of tensorflow/models@e006f5f0d5 (2026-08-24). Data as JSON: /api/errors/c297bf736e115daf. Report an issue: GitHub.