tensorflow/models · error · ValueError

use_causal_windowed and short_seq methods are mutually exclu

Error message

use_causal_windowed and short_seq methods are mutually exclusive

What it means

Error "use_causal_windowed and short_seq methods are mutually exclusive" thrown in tensorflow/models.

Source

Thrown at official/nlp/modeling/layers/kernel_attention.py:633

    # 2. no redraw
    self._seed = seed
    super().__init__(**kwargs)
    if scale is None:
      self._scale = 1.0 / math.sqrt(float(self._key_dim))
    else:
      self._scale = scale
    self._projection_matrix = None
    if num_random_features > 0:
      self._projection_matrix = create_projection_matrix(
          self._num_random_features, self._key_dim,
          tf.constant([self._seed, self._seed + 1]))
    self.use_causal_windowed = use_causal_windowed
    self.causal_chunk_length = causal_chunk_length
    self.causal_window_length = causal_window_length
    self.causal_window_decay = causal_window_decay
    self.causal_padding = causal_padding
    if self.use_causal_windowed and self._is_short_seq:
      raise ValueError(
          "use_causal_windowed and short_seq methods are mutually exclusive")

  def _compute_attention(self,
                         query,
                         key,
                         value,
                         feature_transform,
                         is_short_seq,
                         attention_mask=None,
                         cache=None,
                         training=False,
                         numeric_stabler=_NUMERIC_STABLER):
    """Applies kernel attention with query, key, value tensors.

    This function defines the computation inside `call` with projected
    multi-head Q, K, V inputs. Users can override this function for customized
    attention implementation.

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/nlp/modeling/layers/kernel_attention.py:633 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/54b2bc59905c4594. Report an issue: GitHub.