[Docs] Fix `WhisperEncoderLayer.forward` docstring in `dots3_note` (#52134)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Aug 13, 2026 at 10:32 UTC
903da602f3868a10aa418ba5513c280dfa7ef635
1 file changed
+18
-9
vllm/models/dots3_note/nvidia/audio_encoder.py
+18
-9
@@ -345,17 +345,26 @@ class WhisperEncoderLayer(nn.Module):
345
output_attentions: bool = False,
346
rotary_cos: torch.Tensor | None = None,
347
rotary_sin: torch.Tensor | None = None,
348
- ) -> torch.Tensor:
348
+ ) -> tuple[Any, ...]:
349
"""
350
Args:
351
- hidden_states (`torch.FloatTensor`): input to the layer of shape `(seq_len, batch, embed_dim)`
352
- attention_mask (`torch.FloatTensor`): attention mask of size
353
- `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
354
- layer_head_mask (`torch.FloatTensor`): mask for attention heads in a given layer of size
355
- `(encoder_attention_heads,)`.
356
- output_attentions (`bool`, *optional*):
357
- Whether or not to return the attentions tensors of all attention layers. See `attentions` under
358
- returned tensors for more detail.
351
+ hidden_states: Input to the layer of shape `(batch, seq_len,
352
+ embed_dim)`, or `(total_tokens, embed_dim)` when `cu_seqlens_q`
353
+ is given.
354
+ cu_seqlens_q: Cumulative query sequence lengths for packed
355
+ variable-length input. If `None`, the input is treated as a
356
+ padded batch.
357
+ cu_seqlens_kv: Cumulative key/value sequence lengths for packed
358
+ variable-length input.
359
+ max_seqlen_q: Longest query sequence in the packed batch.
360
+ max_seqlen_kv: Longest key/value sequence in the packed batch.
361
+ output_attentions: Whether to also return the attention weights.
362
+ rotary_cos: Cosine component of the rotary position embedding.
363
+ rotary_sin: Sine component of the rotary position embedding.
364
+
365
+ Returns:
366
+ A tuple of the output hidden states, followed by the attention
367
+ weights if `output_attentions` is `True`.
368
"""
369
residual = hidden_states
370
hidden_states = self.self_attn_layer_norm(hidden_states)