shorten comment

src/audiocraft/modules/transformer.py

@@ -399,9 +399,8 @@ class StreamingMultiheadAttention(StreamingModule):
                     v = expand_repeated_kv(v, self.kv_repeat, self.memory_efficient)
             if self.attention_as_float32:
                 q, k, v = [x.float() for x in [q, k, v]]
-            # Original: this block was only reachable inside `if self.memory_efficient:`.
-            # Moved out so the plain (non-memory-efficient) path below also gets the
-            # align8-padding trim -- it needs the same seq_len match as the mask.
+            # block was initially reachable inside 'if self.memory_efficient'
+            # moved out so plain attn also gets trimmed to match seq_len's mask.
             if custom_attn_mask:
                 # When using a custom attn mask:
                 # Move to query's device, repeat for each sample, remove align8 padding