transformers 5.0.0rc1__py3-none-any.whl → 5.0.0rc3__py3-none-any.whl

transformers/models/git/processing_git.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2022 The HuggingFace Inc. team.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -17,22 +16,11 @@ Image/Text processor class for GIT
 """
 
 from ...processing_utils import ProcessorMixin
+from ...utils import auto_docstring
 
 
+@auto_docstring
 class GitProcessor(ProcessorMixin):
-    r"""
-    Constructs a GIT processor which wraps a CLIP image processor and a BERT tokenizer into a single processor.
-
-    [`GitProcessor`] offers all the functionalities of [`CLIPImageProcessor`] and [`BertTokenizerFast`]. See the
-    [`~GitProcessor.__call__`] and [`~GitProcessor.decode`] for more information.
-
-    Args:
-        image_processor ([`AutoImageProcessor`]):
-            The image processor is a required input.
-        tokenizer ([`AutoTokenizer`]):
-            The tokenizer is a required input.
-    """
-
     def __init__(self, image_processor, tokenizer):
         super().__init__(image_processor, tokenizer)
 

transformers/models/glm/configuration_glm.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 The GLM & ZhipuAI team and HuggingFace Inc. team. All rights reserved.
 #
 #
@@ -14,7 +13,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Optional
 
 from ...configuration_utils import PreTrainedConfig
 from ...modeling_rope_utils import RopeParameters
@@ -105,25 +103,25 @@ class GlmConfig(PreTrainedConfig):
 
     def __init__(
         self,
-        vocab_size: Optional[int] = 151552,
-        hidden_size: Optional[int] = 4096,
-        intermediate_size: Optional[int] = 13696,
-        num_hidden_layers: Optional[int] = 40,
-        num_attention_heads: Optional[int] = 32,
-        num_key_value_heads: Optional[int] = 2,
-        head_dim: Optional[int] = 128,
-        hidden_act: Optional[str] = "silu",
-        attention_dropout: Optional[float] = 0.0,
-        max_position_embeddings: Optional[int] = 131072,
-        initializer_range: Optional[float] = 0.02,
-        rms_norm_eps: Optional[float] = 0.00000015625,
-        use_cache: Optional[bool] = True,
-        tie_word_embeddings: Optional[bool] = False,
-        rope_parameters: Optional[RopeParameters | dict[str, RopeParameters]] = None,
-        pad_token_id: Optional[int] = 151329,
-        eos_token_id: Optional[list[int]] = [151329, 151336, 151338],
-        bos_token_id: Optional[int] = None,
-        attention_bias: Optional[bool] = True,
+        vocab_size: int | None = 151552,
+        hidden_size: int | None = 4096,
+        intermediate_size: int | None = 13696,
+        num_hidden_layers: int | None = 40,
+        num_attention_heads: int | None = 32,
+        num_key_value_heads: int | None = 2,
+        head_dim: int | None = 128,
+        hidden_act: str | None = "silu",
+        attention_dropout: float | None = 0.0,
+        max_position_embeddings: int | None = 131072,
+        initializer_range: float | None = 0.02,
+        rms_norm_eps: float | None = 0.00000015625,
+        use_cache: bool | None = True,
+        tie_word_embeddings: bool | None = False,
+        rope_parameters: RopeParameters | dict[str, RopeParameters] | None = None,
+        pad_token_id: int | None = 151329,
+        eos_token_id: list[int] | None = [151329, 151336, 151338],
+        bos_token_id: int | None = None,
+        attention_bias: bool | None = True,
         **kwargs,
     ):
         self.vocab_size = vocab_size

transformers/models/glm/modeling_glm.py

@@ -4,7 +4,6 @@
 #             the file from the modular. If any change should be done, please apply the change to the
 #                          modular_glm.py file directly. One of our CI enforces this.
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
-# coding=utf-8
 # Copyright 2024 The GLM & ZhipuAI team and HuggingFace Inc. team. All rights reserved.
 #
 #
@@ -20,7 +19,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from collections.abc import Callable
-from typing import Optional, Union
+from typing import Optional
 
 import torch
 import torch.nn as nn
@@ -79,13 +78,13 @@ class GlmRotaryEmbedding(nn.Module):
         inv_freq, self.attention_scaling = rope_init_fn(self.config, device)
 
         self.register_buffer("inv_freq", inv_freq, persistent=False)
-        self.original_inv_freq = inv_freq
+        self.register_buffer("original_inv_freq", inv_freq.clone(), persistent=False)
 
     @staticmethod
     def compute_default_rope_parameters(
-        config: Optional[GlmConfig] = None,
+        config: GlmConfig | 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
@@ -146,7 +145,7 @@ 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],
@@ -174,7 +173,7 @@ def rotate_half(x):
     return torch.stack((-x2, x1), dim=-1).flatten(-2)
 
 
-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:
@@ -182,8 +181,6 @@ 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
@@ -220,7 +217,7 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
 class GlmAttention(nn.Module):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
-    def __init__(self, config: GlmConfig, layer_idx: Optional[int] = None):
+    def __init__(self, config: GlmConfig, layer_idx: int | None = None):
         super().__init__()
         self.config = config
         self.layer_idx = layer_idx
@@ -244,10 +241,10 @@ class GlmAttention(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
-        attention_mask: Optional[torch.Tensor] = None,
-        past_key_values: Optional[Cache] = None,
-        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        attention_mask: torch.Tensor | None = None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> tuple[torch.Tensor, torch.Tensor]:
         input_shape = hidden_states.shape[:-1]
@@ -320,12 +317,12 @@ class GlmDecoderLayer(GradientCheckpointingLayer):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Cache] = None,
-        use_cache: Optional[bool] = False,
-        cache_position: Optional[torch.LongTensor] = None,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        use_cache: bool | None = False,
+        cache_position: torch.LongTensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> torch.Tensor:
         residual = hidden_states
@@ -392,13 +389,13 @@ class GlmModel(GlmPreTrainedModel):
     @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,
-        cache_position: Optional[torch.LongTensor] = None,
-        use_cache: Optional[bool] = 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,
+        cache_position: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> BaseModelOutputWithPast:
         if (input_ids is None) ^ (inputs_embeds is not None):
@@ -469,15 +466,15 @@ class GlmForCausalLM(GlmPreTrainedModel, 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,
-        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,
+        cache_position: torch.LongTensor | None = None,
+        logits_to_keep: int | torch.Tensor = 0,
         **kwargs: Unpack[TransformersKwargs],
     ) -> CausalLMOutputWithPast:
         r"""

transformers/models/glm/modular_glm.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 The GLM & ZhipuAI team and HuggingFace Inc. team. All rights reserved.
 #
 #
@@ -42,9 +41,9 @@ class GlmMLP(Phi3MLP):
 class GlmRotaryEmbedding(LlamaRotaryEmbedding):
     @staticmethod
     def compute_default_rope_parameters(
-        config: Optional[GlmConfig] = None,
+        config: GlmConfig | 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
@@ -80,7 +79,7 @@ def rotate_half(x):
     return torch.stack((-x2, x1), dim=-1).flatten(-2)
 
 
-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:
@@ -88,8 +87,6 @@ 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
@@ -123,7 +120,7 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
 
 
 class GlmAttention(LlamaAttention):
-    def __init__(self, config: GlmConfig, layer_idx: Optional[int] = None):
+    def __init__(self, config: GlmConfig, layer_idx: int | None = None):
         super().__init__(config, layer_idx)
         self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
 

transformers/models/glm4/configuration_glm4.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2025 The GLM4 & ZhipuAI team and HuggingFace Inc. team. All rights reserved.
 #
 #
@@ -14,7 +13,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Optional
 
 from ...configuration_utils import PreTrainedConfig
 from ...modeling_rope_utils import RopeParameters
@@ -105,25 +103,25 @@ class Glm4Config(PreTrainedConfig):
 
     def __init__(
         self,
-        vocab_size: Optional[int] = 151552,
-        hidden_size: Optional[int] = 4096,
-        intermediate_size: Optional[int] = 13696,
-        num_hidden_layers: Optional[int] = 40,
-        num_attention_heads: Optional[int] = 32,
-        num_key_value_heads: Optional[int] = 2,
-        head_dim: Optional[int] = 128,
-        hidden_act: Optional[str] = "silu",
-        attention_dropout: Optional[float] = 0.0,
-        max_position_embeddings: Optional[int] = 131072,
-        initializer_range: Optional[float] = 0.02,
-        rms_norm_eps: Optional[float] = 0.00000015625,
-        use_cache: Optional[bool] = True,
-        tie_word_embeddings: Optional[bool] = False,
-        rope_parameters: Optional[RopeParameters | dict[str, RopeParameters]] = None,
-        pad_token_id: Optional[int] = 151329,
-        eos_token_id: Optional[list[int]] = [151329, 151336, 151338],
-        bos_token_id: Optional[int] = None,
-        attention_bias: Optional[bool] = True,
+        vocab_size: int | None = 151552,
+        hidden_size: int | None = 4096,
+        intermediate_size: int | None = 13696,
+        num_hidden_layers: int | None = 40,
+        num_attention_heads: int | None = 32,
+        num_key_value_heads: int | None = 2,
+        head_dim: int | None = 128,
+        hidden_act: str | None = "silu",
+        attention_dropout: float | None = 0.0,
+        max_position_embeddings: int | None = 131072,
+        initializer_range: float | None = 0.02,
+        rms_norm_eps: float | None = 0.00000015625,
+        use_cache: bool | None = True,
+        tie_word_embeddings: bool | None = False,
+        rope_parameters: RopeParameters | dict[str, RopeParameters] | None = None,
+        pad_token_id: int | None = 151329,
+        eos_token_id: list[int] | None = [151329, 151336, 151338],
+        bos_token_id: int | None = None,
+        attention_bias: bool | None = True,
         **kwargs,
     ):
         self.vocab_size = vocab_size

transformers/models/glm4/modeling_glm4.py

@@ -4,7 +4,6 @@
 #             the file from the modular. If any change should be done, please apply the change to the
 #                          modular_glm4.py file directly. One of our CI enforces this.
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
-# coding=utf-8
 # Copyright 2025 The GLM4 & ZhipuAI team and HuggingFace Inc. team. All rights reserved.
 #
 #
@@ -19,8 +18,9 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 from collections.abc import Callable
-from typing import Optional, Union
+from typing import Optional
 
 import torch
 import torch.nn as nn
@@ -78,14 +78,14 @@ class Glm4DecoderLayer(GradientCheckpointingLayer):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Cache] = None,
-        use_cache: Optional[bool] = False,
-        cache_position: Optional[torch.LongTensor] = None,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        use_cache: bool | None = False,
+        cache_position: torch.LongTensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
         **kwargs: Unpack[FlashAttentionKwargs],
-    ) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
+    ) -> tuple[torch.FloatTensor, tuple[torch.FloatTensor, torch.FloatTensor] | None]:
         residual = hidden_states
         hidden_states = self.input_layernorm(hidden_states)
 
@@ -128,7 +128,7 @@ 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],
@@ -156,7 +156,7 @@ def rotate_half(x):
     return torch.stack((-x2, x1), dim=-1).flatten(-2)
 
 
-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:
@@ -164,8 +164,6 @@ 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
@@ -202,7 +200,7 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
 class Glm4Attention(nn.Module):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
-    def __init__(self, config: Glm4Config, layer_idx: Optional[int] = None):
+    def __init__(self, config: Glm4Config, layer_idx: int | None = None):
         super().__init__()
         self.config = config
         self.layer_idx = layer_idx
@@ -226,10 +224,10 @@ class Glm4Attention(nn.Module):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
-        attention_mask: Optional[torch.Tensor] = None,
-        past_key_values: Optional[Cache] = None,
-        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        attention_mask: torch.Tensor | None = None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> tuple[torch.Tensor, torch.Tensor]:
         input_shape = hidden_states.shape[:-1]
@@ -284,13 +282,13 @@ class Glm4RotaryEmbedding(nn.Module):
         inv_freq, self.attention_scaling = rope_init_fn(self.config, device)
 
         self.register_buffer("inv_freq", inv_freq, persistent=False)
-        self.original_inv_freq = inv_freq
+        self.register_buffer("original_inv_freq", inv_freq.clone(), persistent=False)
 
     @staticmethod
     def compute_default_rope_parameters(
-        config: Optional[Glm4Config] = None,
+        config: Glm4Config | 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
@@ -396,13 +394,13 @@ class Glm4Model(Glm4PreTrainedModel):
     @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,
-        cache_position: Optional[torch.LongTensor] = None,
-        use_cache: Optional[bool] = 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,
+        cache_position: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
         **kwargs: Unpack[TransformersKwargs],
     ) -> BaseModelOutputWithPast:
         if (input_ids is None) ^ (inputs_embeds is not None):
@@ -473,17 +471,17 @@ class Glm4ForCausalLM(Glm4PreTrainedModel, 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,
-        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,
+        cache_position: torch.LongTensor | None = None,
+        logits_to_keep: int | torch.Tensor = 0,
         **kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, CausalLMOutputWithPast]:
+    ) -> tuple | CausalLMOutputWithPast:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
             Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,

transformers/models/glm4/modular_glm4.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2025 The GLM4 & ZhipuAI team and HuggingFace Inc. team. All rights reserved.
 #
 #
@@ -13,7 +12,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-from typing import Optional, Union
 
 import torch
 
@@ -53,14 +51,14 @@ class Glm4DecoderLayer(GradientCheckpointingLayer):
     def forward(
         self,
         hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Cache] = None,
-        use_cache: Optional[bool] = False,
-        cache_position: Optional[torch.LongTensor] = None,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        use_cache: bool | None = False,
+        cache_position: torch.LongTensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
         **kwargs: Unpack[FlashAttentionKwargs],
-    ) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
+    ) -> tuple[torch.FloatTensor, tuple[torch.FloatTensor, torch.FloatTensor] | None]:
         residual = hidden_states
         hidden_states = self.input_layernorm(hidden_states)
 
@@ -94,7 +92,7 @@ class Glm4ForCausalLM(GlmForCausalLM):
     def forward(
         self,
         **super_kwargs: Unpack[TransformersKwargs],
-    ) -> Union[tuple, CausalLMOutputWithPast]:
+    ) -> tuple | CausalLMOutputWithPast:
         r"""
         labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
             Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,

transformers/models/glm46v/configuration_glm46v.py

@@ -4,7 +4,6 @@
 #             the file from the modular. If any change should be done, please apply the change to the
 #                          modular_glm46v.py file directly. One of our CI enforces this.
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
-# coding=utf-8
 # Copyright 2025 the HuggingFace Team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");