transformers 5.0.0rc2__py3-none-any.whl → 5.0.0rc3__py3-none-any.whl

transformers/models/data2vec/modeling_data2vec_vision.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2022 Meta Platforms and The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -17,7 +16,7 @@
 import collections.abc
 import math
 from dataclasses import dataclass
-from typing import Optional, Union
+from typing import Optional
 
 import torch
 from torch import nn
@@ -77,7 +76,7 @@ def drop_path(input: torch.Tensor, drop_prob: float = 0.0, training: bool = Fals
 class Data2VecVisionDropPath(nn.Module):
     """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks)."""
 
-    def __init__(self, drop_prob: Optional[float] = None) -> None:
+    def __init__(self, drop_prob: float | None = None) -> None:
         super().__init__()
         self.drop_prob = drop_prob
 
@@ -161,7 +160,7 @@ class Data2VecVisionEmbeddings(nn.Module):
     def forward(
         self,
         pixel_values: torch.Tensor,
-        bool_masked_pos: Optional[torch.BoolTensor] = None,
+        bool_masked_pos: torch.BoolTensor | None = None,
     ) -> torch.Tensor:
         _, _, height, width = pixel_values.shape
         embeddings, (patch_height, patch_width) = self.patch_embeddings(pixel_values)
@@ -225,7 +224,7 @@ class Data2VecVisionPatchEmbeddings(nn.Module):
 
 # Copied from transformers.models.beit.modeling_beit.BeitSelfAttention with Beit->Data2VecVision
 class Data2VecVisionSelfAttention(nn.Module):
-    def __init__(self, config: Data2VecVisionConfig, window_size: Optional[tuple] = None) -> None:
+    def __init__(self, config: Data2VecVisionConfig, window_size: tuple | None = None) -> None:
         super().__init__()
         self.config = config
         if config.hidden_size % config.num_attention_heads != 0 and not hasattr(config, "embedding_size"):
@@ -252,10 +251,10 @@ class Data2VecVisionSelfAttention(nn.Module):
         self,
         hidden_states: torch.Tensor,
         output_attentions: bool = False,
-        relative_position_bias: Optional[torch.Tensor] = None,
+        relative_position_bias: torch.Tensor | None = None,
         interpolate_pos_encoding: bool = False,
-        resolution: Optional[tuple[int]] = None,
-    ) -> Union[tuple[torch.Tensor], tuple[torch.Tensor, torch.Tensor]]:
+        resolution: tuple[int] | None = None,
+    ) -> tuple[torch.Tensor] | tuple[torch.Tensor, torch.Tensor]:
         batch_size, seq_length, _ = hidden_states.shape
         query_layer = (
             self.query(hidden_states)
@@ -314,10 +313,10 @@ class Data2VecVisionSdpaSelfAttention(Data2VecVisionSelfAttention):
         self,
         hidden_states: torch.Tensor,
         output_attentions: bool = False,
-        relative_position_bias: Optional[torch.Tensor] = None,
+        relative_position_bias: torch.Tensor | None = None,
         interpolate_pos_encoding: bool = False,
-        resolution: Optional[tuple[int]] = None,
-    ) -> Union[tuple[torch.Tensor], tuple[torch.Tensor, torch.Tensor]]:
+        resolution: tuple[int] | None = None,
+    ) -> tuple[torch.Tensor] | tuple[torch.Tensor, torch.Tensor]:
         if output_attentions:
             logger.warning_once(
                 f"{self.__class__.__name__} does not support `output_attentions=True`. The returned attention weights will "
@@ -398,7 +397,7 @@ DATA2VEC_VISION_SELF_ATTENTION_CLASSES = {
 
 # Copied from tests.models.beit.modeling_beit.BeitAttention with Beit->Data2VecVision, BEIT->DATA2VEC_VISION
 class Data2VecVisionAttention(nn.Module):
-    def __init__(self, config: Data2VecVisionConfig, window_size: Optional[tuple] = None) -> None:
+    def __init__(self, config: Data2VecVisionConfig, window_size: tuple | None = None) -> None:
         super().__init__()
         self.attention = DATA2VEC_VISION_SELF_ATTENTION_CLASSES[config._attn_implementation](
             config, window_size=window_size
@@ -411,8 +410,8 @@ class Data2VecVisionAttention(nn.Module):
         output_attentions: bool = False,
         relative_position_bias: Optional["Data2VecVisionRelativePositionBias"] = None,
         interpolate_pos_encoding: bool = False,
-        resolution: Optional[tuple[int]] = None,
-    ) -> Union[tuple[torch.Tensor], tuple[torch.Tensor, torch.Tensor]]:
+        resolution: tuple[int] | None = None,
+    ) -> tuple[torch.Tensor] | tuple[torch.Tensor, torch.Tensor]:
         self_outputs = self.attention(
             hidden_states, output_attentions, relative_position_bias, interpolate_pos_encoding, resolution
         )
@@ -459,7 +458,7 @@ class Data2VecVisionLayer(GradientCheckpointingLayer):
     """This corresponds to the Block class in the timm implementation."""
 
     def __init__(
-        self, config: Data2VecVisionConfig, window_size: Optional[tuple] = None, drop_path_rate: float = 0.0
+        self, config: Data2VecVisionConfig, window_size: tuple | None = None, drop_path_rate: float = 0.0
     ) -> None:
         super().__init__()
         self.chunk_size_feed_forward = config.chunk_size_feed_forward
@@ -482,10 +481,10 @@ class Data2VecVisionLayer(GradientCheckpointingLayer):
         self,
         hidden_states: torch.Tensor,
         output_attentions: bool = False,
-        relative_position_bias: Optional[torch.Tensor] = None,
+        relative_position_bias: torch.Tensor | None = None,
         interpolate_pos_encoding: bool = False,
-        resolution: Optional[tuple[int, int]] = None,
-    ) -> Union[tuple[torch.Tensor], tuple[torch.Tensor, torch.Tensor]]:
+        resolution: tuple[int, int] | None = None,
+    ) -> tuple[torch.Tensor] | tuple[torch.Tensor, torch.Tensor]:
         self_attention_outputs = self.attention(
             self.layernorm_before(hidden_states),  # in Data2VecVision, layernorm is applied before self-attention
             output_attentions=output_attentions,
@@ -606,7 +605,7 @@ class Data2VecVisionRelativePositionBias(nn.Module):
 
 # Copied from transformers.models.beit.modeling_beit.BeitEncoder with Beit->Data2VecVision
 class Data2VecVisionEncoder(nn.Module):
-    def __init__(self, config: Data2VecVisionConfig, window_size: Optional[tuple] = None) -> None:
+    def __init__(self, config: Data2VecVisionConfig, window_size: tuple | None = None) -> None:
         super().__init__()
         self.config = config
         self.has_relative_position_bias = config.use_shared_relative_position_bias
@@ -633,9 +632,9 @@ class Data2VecVisionEncoder(nn.Module):
         output_attentions: bool = False,
         output_hidden_states: bool = False,
         interpolate_pos_encoding: bool = False,
-        resolution: Optional[tuple[int, int]] = None,
+        resolution: tuple[int, int] | None = None,
         return_dict: bool = True,
-    ) -> Union[tuple, BaseModelOutput]:
+    ) -> tuple | BaseModelOutput:
         all_hidden_states = () if output_hidden_states else None
         all_self_attentions = () if output_attentions else None
 
@@ -736,13 +735,13 @@ class Data2VecVisionModel(Data2VecVisionPreTrainedModel):
     def forward(
         self,
         pixel_values: torch.Tensor,
-        bool_masked_pos: Optional[torch.BoolTensor] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
+        bool_masked_pos: torch.BoolTensor | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
         interpolate_pos_encoding: bool = False,
-        return_dict: Optional[bool] = None,
+        return_dict: bool | None = None,
         **kwargs,
-    ) -> Union[tuple, Data2VecVisionModelOutputWithPooling]:
+    ) -> tuple | Data2VecVisionModelOutputWithPooling:
         r"""
         bool_masked_pos (`torch.BoolTensor` of shape `(batch_size, num_patches)`, *optional*):
             Boolean masked positions. Indicates which patches are masked (1) and which aren't (0).
@@ -823,14 +822,14 @@ class Data2VecVisionForImageClassification(Data2VecVisionPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        pixel_values: Optional[torch.Tensor] = None,
-        labels: Optional[torch.Tensor] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
+        pixel_values: torch.Tensor | None = None,
+        labels: torch.Tensor | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
         interpolate_pos_encoding: bool = False,
-        return_dict: Optional[bool] = None,
+        return_dict: bool | None = None,
         **kwargs,
-    ) -> Union[tuple, ImageClassifierOutput]:
+    ) -> tuple | ImageClassifierOutput:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the image classification/regression loss. Indices should be in `[0, ...,
@@ -879,10 +878,10 @@ class Data2VecVisionConvModule(nn.Module):
         self,
         in_channels: int,
         out_channels: int,
-        kernel_size: Union[int, tuple[int, int]],
-        padding: Union[int, tuple[int, int], str] = 0,
+        kernel_size: int | tuple[int, int],
+        padding: int | tuple[int, int] | str = 0,
         bias: bool = False,
-        dilation: Union[int, tuple[int, int]] = 1,
+        dilation: int | tuple[int, int] = 1,
     ) -> None:
         super().__init__()
         self.conv = nn.Conv2d(
@@ -1069,7 +1068,7 @@ class Data2VecVisionFCNHead(nn.Module):
         config: Data2VecVisionConfig,
         in_index: int = 2,
         kernel_size: int = 3,
-        dilation: Union[int, tuple[int, int]] = 1,
+        dilation: int | tuple[int, int] = 1,
     ) -> None:
         super().__init__()
         self.in_channels = config.hidden_size
@@ -1169,14 +1168,14 @@ class Data2VecVisionForSemanticSegmentation(Data2VecVisionPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        pixel_values: Optional[torch.Tensor] = None,
-        labels: Optional[torch.Tensor] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
+        pixel_values: torch.Tensor | None = None,
+        labels: torch.Tensor | None = None,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
         interpolate_pos_encoding: bool = False,
-        return_dict: Optional[bool] = None,
+        return_dict: bool | None = None,
         **kwargs,
-    ) -> Union[tuple, SemanticSegmenterOutput]:
+    ) -> tuple | SemanticSegmenterOutput:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, height, width)`, *optional*):
             Ground truth semantic segmentation maps for computing the loss. Indices should be in `[0, ...,

transformers/models/data2vec/modular_data2vec_audio.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2022 The HuggingFace Inc. team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -215,6 +214,12 @@ class Data2VecAudioModel(Data2VecAudioPreTrainedModel, Wav2Vec2Model):
 
 class Data2VecAudioForCTC(Data2VecAudioPreTrainedModel, Wav2Vec2ForCTC):
     def __init__(self, config):
+        r"""
+        config ([`Data2VecAudioForCTC`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`]  method to load the model weights.
+        """
         Data2VecAudioPreTrainedModel.__init__(self, config)
 
         self.data2vec_audio = Data2VecAudioModel(config)

transformers/models/data2vec/modular_data2vec_text.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2022 The HuggingFace Inc. team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -14,8 +13,6 @@
 # limitations under the License.
 """PyTorch Data2VecText model."""
 
-from typing import Optional, Union
-
 import torch
 import torch.nn as nn
 from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
@@ -135,20 +132,20 @@ class Data2VecTextForCausalLM(Data2VecTextPreTrainedModel, GenerationMixin):
     @auto_docstring
     def forward(
         self,
-        input_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        encoder_attention_mask: Optional[torch.FloatTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[tuple[tuple[torch.FloatTensor]]] = None,
-        use_cache: Optional[bool] = None,
-        cache_position: Optional[torch.Tensor] = None,
-        logits_to_keep: Union[int, torch.Tensor] = 0,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.FloatTensor | None = None,
+        token_type_ids: torch.LongTensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        encoder_hidden_states: torch.FloatTensor | None = None,
+        encoder_attention_mask: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
+        past_key_values: tuple[tuple[torch.FloatTensor]] | None = None,
+        use_cache: bool | None = None,
+        cache_position: torch.Tensor | None = None,
+        logits_to_keep: int | torch.Tensor = 0,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, CausalLMOutputWithCrossAttentions]:
+    ) -> tuple | CausalLMOutputWithCrossAttentions:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
             Labels for computing the left-to-right language modeling loss (next word prediction). Indices should be in
@@ -240,16 +237,16 @@ class Data2VecTextForMaskedLM(Data2VecTextPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        input_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        encoder_hidden_states: Optional[torch.FloatTensor] = None,
-        encoder_attention_mask: Optional[torch.FloatTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.FloatTensor | None = None,
+        token_type_ids: torch.LongTensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        encoder_hidden_states: torch.FloatTensor | None = None,
+        encoder_attention_mask: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, MaskedLMOutput]:
+    ) -> tuple | MaskedLMOutput:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
             Labels for computing the masked language modeling loss. Indices should be in `[-100, 0, ...,
@@ -307,14 +304,14 @@ class Data2VecTextForSequenceClassification(Data2VecTextPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        input_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.FloatTensor | None = None,
+        token_type_ids: torch.LongTensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, SequenceClassifierOutput]:
+    ) -> tuple | SequenceClassifierOutput:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
             Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
@@ -382,14 +379,14 @@ class Data2VecTextForMultipleChoice(Data2VecTextPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        input_ids: Optional[torch.LongTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
+        input_ids: torch.LongTensor | None = None,
+        token_type_ids: torch.LongTensor | None = None,
+        attention_mask: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, MultipleChoiceModelOutput]:
+    ) -> tuple | MultipleChoiceModelOutput:
         r"""
         input_ids (`torch.LongTensor` of shape `(batch_size, num_choices, sequence_length)`):
             Indices of input sequence tokens in the vocabulary.
@@ -482,14 +479,14 @@ class Data2VecTextForTokenClassification(Data2VecTextPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        input_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.FloatTensor | None = None,
+        token_type_ids: torch.LongTensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, TokenClassifierOutput]:
+    ) -> tuple | TokenClassifierOutput:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
             Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
@@ -540,15 +537,15 @@ class Data2VecTextForQuestionAnswering(Data2VecTextPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        input_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        start_positions: Optional[torch.LongTensor] = None,
-        end_positions: Optional[torch.LongTensor] = None,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.FloatTensor | None = None,
+        token_type_ids: torch.LongTensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        start_positions: torch.LongTensor | None = None,
+        end_positions: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, QuestionAnsweringModelOutput]:
+    ) -> tuple | QuestionAnsweringModelOutput:
         outputs = self.data2vec_text(
             input_ids,
             attention_mask=attention_mask,

transformers/models/dbrx/configuration_dbrx.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 Databricks Mosaic Research and The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -14,7 +13,7 @@
 # limitations under the License.
 """DBRX model configuration"""
 
-from typing import Any, Optional
+from typing import Any
 
 from ...configuration_utils import PreTrainedConfig
 from ...modeling_rope_utils import RopeParameters
@@ -47,7 +46,7 @@ class DbrxAttentionConfig(PreTrainedConfig):
     def __init__(
         self,
         attn_pdrop: float = 0.0,
-        clip_qkv: Optional[float] = None,
+        clip_qkv: float | None = None,
         kv_n_heads: int = 1,
         **kwargs: Any,
     ):
@@ -83,13 +82,13 @@ class DbrxFFNConfig(PreTrainedConfig):
     def __init__(
         self,
         hidden_size=6144,
-        ffn_act_fn: Optional[dict] = None,
+        ffn_act_fn: dict | None = None,
         ffn_hidden_size: int = 3584,
         moe_num_experts: int = 4,
         moe_top_k: int = 1,
-        moe_jitter_eps: Optional[float] = None,
+        moe_jitter_eps: float | None = None,
         moe_loss_weight: float = 0.01,
-        moe_normalize_expert_weights: Optional[float] = 1.0,
+        moe_normalize_expert_weights: float | None = 1.0,
         **kwargs: Any,
     ):
         super().__init__()
@@ -185,19 +184,19 @@ class DbrxConfig(PreTrainedConfig):
 
     def __init__(
         self,
-        d_model: Optional[int] = 2048,
-        n_heads: Optional[int] = 16,
-        n_layers: Optional[int] = 24,
-        max_seq_len: Optional[int] = 2048,
-        vocab_size: Optional[int] = 32000,
-        resid_pdrop: Optional[float] = 0.0,
-        emb_pdrop: Optional[float] = 0.0,
-        attn_config: Optional[DbrxAttentionConfig] = None,
-        ffn_config: Optional[DbrxFFNConfig] = None,
-        use_cache: Optional[bool] = True,
-        initializer_range: Optional[float] = 0.02,
-        output_router_logits: Optional[bool] = False,
-        rope_parameters: Optional[RopeParameters | dict[str, RopeParameters]] = None,
+        d_model: int | None = 2048,
+        n_heads: int | None = 16,
+        n_layers: int | None = 24,
+        max_seq_len: int | None = 2048,
+        vocab_size: int | None = 32000,
+        resid_pdrop: float | None = 0.0,
+        emb_pdrop: float | None = 0.0,
+        attn_config: DbrxAttentionConfig | None = None,
+        ffn_config: DbrxFFNConfig | None = None,
+        use_cache: bool | None = True,
+        initializer_range: float | None = 0.02,
+        output_router_logits: bool | None = False,
+        rope_parameters: RopeParameters | dict[str, RopeParameters] | None = None,
         **kwargs: Any,
     ):
         if attn_config is None:

transformers/models/dbrx/modeling_dbrx.py

@@ -4,7 +4,6 @@
 #             the file from the modular. If any change should be done, please apply the change to the
 #                          modular_dbrx.py file directly. One of our CI enforces this.
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
-# coding=utf-8
 # Copyright 2024 Databricks Mosaic Research and The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -20,7 +19,7 @@
 # limitations under the License.
 
 from collections.abc import Callable
-from typing import Any, Optional, Union
+from typing import Any, Optional
 
 import torch
 from torch import nn
@@ -62,9 +61,9 @@ class DbrxRotaryEmbedding(nn.Module):
 
     @staticmethod
     def compute_default_rope_parameters(
-        config: Optional[DbrxConfig] = None,
+        config: DbrxConfig | None = None,
         device: Optional["torch.device"] = None,
-        seq_len: Optional[int] = None,
+        seq_len: int | None = None,
     ) -> tuple["torch.Tensor", float]:
         """
         Computes the inverse frequencies according to the original RoPE implementation
@@ -114,7 +113,7 @@ def rotate_half(x):
 
 
 @use_kernel_func_from_hub("rotary_pos_emb")
-def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1):
     """Applies Rotary Position Embedding to the query and key tensors.
 
     Args:
@@ -122,8 +121,6 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
         k (`torch.Tensor`): The key tensor.
         cos (`torch.Tensor`): The cosine part of the rotary embedding.
         sin (`torch.Tensor`): The sine part of the rotary embedding.
-        position_ids (`torch.Tensor`, *optional*):
-            Deprecated and unused.
         unsqueeze_dim (`int`, *optional*, defaults to 1):
             The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
             sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
@@ -158,7 +155,7 @@ def eager_attention_forward(
     query: torch.Tensor,
     key: torch.Tensor,
     value: torch.Tensor,
-    attention_mask: Optional[torch.Tensor],
+    attention_mask: torch.Tensor | None,
     scaling: float,
     dropout: float = 0.0,
     **kwargs: Unpack[TransformersKwargs],
@@ -185,7 +182,7 @@ class DbrxAttention(nn.Module):
     def __init__(
         self,
         config,
-        layer_idx: Optional[int] = None,
+        layer_idx: int | None = None,
         **kwargs,
     ):
         super().__init__()
@@ -213,10 +210,10 @@ class DbrxAttention(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_embeddings: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Cache] = None,
-        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: torch.Tensor | None = None,
+        position_embeddings: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         input_shape = hidden_states.shape[:-1]
@@ -376,7 +373,7 @@ class DbrxFFN(nn.Module):
 
 
 class DbrxNormAttentionNorm(nn.Module):
-    def __init__(self, config: DbrxConfig, layer_idx: Optional[int] = None):
+    def __init__(self, config: DbrxConfig, layer_idx: int | None = None):
         super().__init__()
         self.layer_idx = layer_idx
         self.resid_pdrop = config.resid_pdrop
@@ -391,9 +388,9 @@ class DbrxNormAttentionNorm(nn.Module):
         self,
         hidden_states: torch.Tensor,
         position_embeddings: torch.LongTensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        past_key_values: Optional[Cache] = None,
-        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: torch.Tensor | None = None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs: Any,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         residual_states = hidden_states
@@ -432,10 +429,10 @@ class DbrxBlock(GradientCheckpointingLayer):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_embeddings: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Cache] = None,
-        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: torch.Tensor | None = None,
+        position_embeddings: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs: Any,
     ):
         resid_states, hidden_states = self.norm_attn_norm(
@@ -513,13 +510,13 @@ class DbrxModel(DbrxPreTrainedModel):
     @auto_docstring
     def forward(
         self,
-        input_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Cache] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        use_cache: Optional[bool] = None,
-        cache_position: Optional[torch.LongTensor] = None,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        use_cache: bool | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> MoeModelOutputWithPast:
         if (input_ids is None) ^ (inputs_embeds is not None):
@@ -574,11 +571,11 @@ class DbrxModel(DbrxPreTrainedModel):
 
 
 def load_balancing_loss_func(
-    gate_logits: Union[torch.Tensor, tuple[torch.Tensor], None],
-    num_experts: Optional[int] = None,
+    gate_logits: torch.Tensor | tuple[torch.Tensor] | None,
+    num_experts: int | None = None,
     top_k=2,
-    attention_mask: Optional[torch.Tensor] = None,
-) -> Union[torch.Tensor, int]:
+    attention_mask: torch.Tensor | None = None,
+) -> torch.Tensor | int:
     r"""
     Computes auxiliary load balancing loss as in Switch Transformer - implemented in Pytorch.
 
@@ -692,16 +689,16 @@ class DbrxForCausalLM(DbrxPreTrainedModel, GenerationMixin):
     @auto_docstring
     def forward(
         self,
-        input_ids: Optional[torch.LongTensor] = None,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Cache] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
-        use_cache: Optional[bool] = None,
-        output_router_logits: Optional[bool] = None,
-        cache_position: Optional[torch.LongTensor] = None,
-        logits_to_keep: Union[int, torch.Tensor] = 0,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
+        output_router_logits: bool | None = None,
+        cache_position: torch.LongTensor | None = None,
+        logits_to_keep: int | torch.Tensor = 0,
         **kwargs: Unpack[TransformersKwargs],
     ) -> MoeCausalLMOutputWithPast:
         r"""