tensorflow/models · error · ValueError

`momentum` must be between [0, 1].

Error message

`momentum` must be between [0, 1].

What it means

Range check in the SGD constructor: the momentum hyper-parameter must be within [0, 1]. Values outside that range (negative or > 1) produce diverging or nonsensical velocity updates in the momentum accumulation, so __init__ rejects them immediately. Fires when momentum comes from a config typo (e.g. 10 instead of 0.10) or a schedule/callable resolving to an out-of-range scalar.

Solutions

1. Set momentum to a value in [0, 1] — typical choices are 0.9 or 0.99
2. Check the YAML/flags for decimal errors (10 vs 0.10)
3. If momentum is a schedule or callable, verify it never yields values outside [0, 1]

When it happens

Trigger: Thrown at official/projects/yolo/optimization/sgd_torch.py:107 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/eb2a913690368a9e. Report an issue: GitHub.

Appendix: source

Thrown at official/projects/yolo/optimization/sgd_torch.py:107

    # Create Hyper Params for each group of the LR
    self._set_hyper("learning_rate", kwargs.get("lr", learning_rate))
    self._set_hyper("bias_learning_rate", kwargs.get("lr", learning_rate))
    self._set_hyper("other_learning_rate", kwargs.get("lr", learning_rate))

    # SGD decay param
    self._set_hyper("decay", self._initial_decay)

    # Weight decay param
    self._weight_decay = weight_decay != 0.0
    self._set_hyper("weight_decay", weight_decay)

    # Enable Momentum
    self._momentum = False
    if isinstance(momentum, tf.Tensor) or callable(momentum) or momentum > 0:
      self._momentum = True
    if isinstance(momentum, (int, float)) and (momentum < 0 or momentum > 1):
      raise ValueError("`momentum` must be between [0, 1].")
    self._set_hyper("momentum", momentum)
    self._set_hyper("momentum_start", momentum_start)
    self._set_hyper("warmup_steps", tf.cast(warmup_steps, tf.int32))

    # Enable Nesterov Momentum
    self.nesterov = nesterov

    # weights, biases, other
    self._weight_keys = weight_keys
    self._bias_keys = bias_keys
    self._variables_set = False
    self._wset = set()
    self._bset = set()
    self._oset = set()

    logging.info("Pytorch SGD simulation: ")
    logging.info("Weight Decay: %f", weight_decay)

View on GitHub (pinned to e006f5f0d5)