nmn 0.1.7__py3-none-any.whl → 0.1.8__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
nmn/nnx/TODO ADDED
@@ -0,0 +1,2 @@
1
+ - add support to masked kernels
2
+ - explain attention [directed graph]
nmn/nnx/yatattention.py CHANGED
@@ -460,8 +460,6 @@ class MultiHeadAttention(Module):
460
460
  self.key_ln = None
461
461
 
462
462
  # Remove the output layer - no more self.out
463
- self.rngs = rngs if dropout_rate > 0.0 else None
464
-
465
463
  self.cached_key: nnx.Cache[Array] | None = None
466
464
  self.cached_value: nnx.Cache[Array] | None = None
467
465
  self.cache_index: nnx.Cache[Array] | None = None
@@ -507,8 +505,6 @@ class MultiHeadAttention(Module):
507
505
  Returns:
508
506
  output of shape `[batch_sizes..., length, features]`.
509
507
  """
510
- if rngs is None:
511
- rngs = self.rngs
512
508
 
513
509
  if inputs_k is None:
514
510
  if inputs_v is not None: