Merge branch 'main' into the-great-cleaning

Author: Cyrilvallez

src/transformers/models/qwen3_vl/modeling_qwen3_vl.py

@@ -528,8 +528,7 @@ def forward(
 class Qwen3VLModelOutputWithPast(ModelOutput):
     r"""
     past_key_values (`Cache`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
-        Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
-        `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+        It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache).
 
         Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
         `past_key_values` input) to speed up sequential decoding.
@@ -538,7 +537,7 @@ class Qwen3VLModelOutputWithPast(ModelOutput):
     """
 
     last_hidden_state: Optional[torch.FloatTensor] = None
-    past_key_values: Optional[list[torch.FloatTensor]] = None
+    past_key_values: Optional[Cache] = None
     hidden_states: Optional[tuple[torch.FloatTensor]] = None
     attentions: Optional[tuple[torch.FloatTensor]] = None
     rope_deltas: Optional[torch.LongTensor] = None
@@ -1255,8 +1254,7 @@ class Qwen3VLCausalLMOutputWithPast(ModelOutput):
     logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
         Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
     past_key_values (`Cache`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
-        Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
-        `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+        It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache).
 
         Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
         `past_key_values` input) to speed up sequential decoding.
@@ -1266,7 +1264,7 @@ class Qwen3VLCausalLMOutputWithPast(ModelOutput):
 
     loss: Optional[torch.FloatTensor] = None
     logits: Optional[torch.FloatTensor] = None
-    past_key_values: Optional[list[torch.FloatTensor]] = None
+    past_key_values: Optional[Cache] = None
     hidden_states: Optional[tuple[torch.FloatTensor]] = None
     attentions: Optional[tuple[torch.FloatTensor]] = None
     rope_deltas: Optional[torch.LongTensor] = None
@@ -1322,7 +1320,7 @@ def forward(
         input_ids: torch.LongTensor = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[list[torch.FloatTensor]] = None,
+        past_key_values: Optional[Cache] = None,
         inputs_embeds: Optional[torch.FloatTensor] = None,
         labels: Optional[torch.LongTensor] = None,
         pixel_values: Optional[torch.Tensor] = None,

src/transformers/models/qwen3_vl/modular_qwen3_vl.py

@@ -1156,7 +1156,7 @@ def forward(
         input_ids: torch.LongTensor = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[list[torch.FloatTensor]] = None,
+        past_key_values: Optional[Cache] = None,
         inputs_embeds: Optional[torch.FloatTensor] = None,
         labels: Optional[torch.LongTensor] = None,
         pixel_values: Optional[torch.Tensor] = None,

src/transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py

@@ -348,7 +348,7 @@ def forward(
         position_embeddings: tuple[torch.Tensor, torch.Tensor],
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[tuple[torch.Tensor]] = None,
+        past_key_values: Optional[Cache] = None,
         cache_position: Optional[torch.LongTensor] = None,
         **kwargs: Unpack[FlashAttentionKwargs],
     ) -> torch.FloatTensor:
@@ -366,7 +366,7 @@ def forward(
             use_cache (`bool`, *optional*):
                 If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
                 (see `past_key_values`).
-            past_key_values (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+            past_key_values (`Cache`, *optional*): cached past key and value projection states
             cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
                 Indices depicting the position of the input sequence tokens in the sequence.
             position_embeddings (`tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*):
@@ -1011,8 +1011,7 @@ def _deepstack_process(
 class Qwen3VLMoeModelOutputWithPast(ModelOutput):
     r"""
     past_key_values (`Cache`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
-        Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
-        `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+        It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache).
 
         Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
         `past_key_values` input) to speed up sequential decoding.
@@ -1021,7 +1020,7 @@ class Qwen3VLMoeModelOutputWithPast(ModelOutput):
     """
 
     last_hidden_state: Optional[torch.FloatTensor] = None
-    past_key_values: Optional[list[torch.FloatTensor]] = None
+    past_key_values: Optional[Cache] = None
     hidden_states: Optional[tuple[torch.FloatTensor]] = None
     attentions: Optional[tuple[torch.FloatTensor]] = None
     rope_deltas: Optional[torch.LongTensor] = None
@@ -1398,8 +1397,7 @@ class Qwen3VLMoeCausalLMOutputWithPast(ModelOutput):
     logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
         Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
     past_key_values (`Cache`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
-        Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
-        `(batch_size, num_heads, sequence_length, embed_size_per_head)`)
+        It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache).
 
         Contains pre-computed hidden-states (key and values in the self-attention blocks) that can be used (see
         `past_key_values` input) to speed up sequential decoding.
@@ -1409,7 +1407,7 @@ class Qwen3VLMoeCausalLMOutputWithPast(ModelOutput):
 
     loss: Optional[torch.FloatTensor] = None
     logits: Optional[torch.FloatTensor] = None
-    past_key_values: Optional[list[torch.FloatTensor]] = None
+    past_key_values: Optional[Cache] = None
     hidden_states: Optional[tuple[torch.FloatTensor]] = None
     attentions: Optional[tuple[torch.FloatTensor]] = None
     rope_deltas: Optional[torch.LongTensor] = None
@@ -1465,7 +1463,7 @@ def forward(
         input_ids: torch.LongTensor = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[list[torch.FloatTensor]] = None,
+        past_key_values: Optional[Cache] = None,
         inputs_embeds: Optional[torch.FloatTensor] = None,
         labels: Optional[torch.LongTensor] = None,
         pixel_values: Optional[torch.Tensor] = None,