transformers 5.0.0rc2__py3-none-any.whl → 5.1.0__py3-none-any.whl

transformers/models/stablelm/configuration_stablelm.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 Stability AI and The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -14,8 +13,6 @@
 # limitations under the License.
 """StableLM model configuration"""
 
-from typing import Optional
-
 from ...configuration_utils import PreTrainedConfig
 from ...modeling_rope_utils import RopeParameters
 from ...utils import logging
@@ -90,6 +87,8 @@ class StableLmConfig(PreTrainedConfig):
             The id of the `BOS` token in the vocabulary.
         eos_token_id (int, *optional*, defaults to 0):
             The id of the `EOS` token in the vocabulary.
+        pad_token_id (int, *optional*):
+            The id of the `PAD` token in the vocabulary.
 
     Example:
 
@@ -105,26 +104,27 @@ class StableLmConfig(PreTrainedConfig):
 
     def __init__(
         self,
-        vocab_size: Optional[int] = 50304,
-        intermediate_size: Optional[int] = 6912,
-        hidden_size: Optional[int] = 2560,
-        num_hidden_layers: Optional[int] = 32,
-        num_attention_heads: Optional[int] = 32,
-        num_key_value_heads: Optional[int] = 32,
-        hidden_act: Optional[str] = "silu",
-        max_position_embeddings: Optional[int] = 4096,
-        initializer_range: Optional[float] = 0.02,
-        layer_norm_eps: Optional[float] = 1.0e-5,
-        use_cache: Optional[bool] = True,
-        tie_word_embeddings: Optional[bool] = False,
-        rope_parameters: Optional[RopeParameters | dict[str, RopeParameters]] = None,
-        use_qkv_bias: Optional[bool] = False,
-        qk_layernorm: Optional[bool] = False,
-        use_parallel_residual: Optional[bool] = False,
-        hidden_dropout: Optional[float] = 0.0,
-        attention_dropout: Optional[float] = 0.0,
-        bos_token_id: Optional[int] = 0,
-        eos_token_id: Optional[int] = 0,
+        vocab_size: int | None = 50304,
+        intermediate_size: int | None = 6912,
+        hidden_size: int | None = 2560,
+        num_hidden_layers: int | None = 32,
+        num_attention_heads: int | None = 32,
+        num_key_value_heads: int | None = 32,
+        hidden_act: str | None = "silu",
+        max_position_embeddings: int | None = 4096,
+        initializer_range: float | None = 0.02,
+        layer_norm_eps: float | None = 1.0e-5,
+        use_cache: bool | None = True,
+        tie_word_embeddings: bool | None = False,
+        rope_parameters: RopeParameters | dict[str, RopeParameters] | None = None,
+        use_qkv_bias: bool | None = False,
+        qk_layernorm: bool | None = False,
+        use_parallel_residual: bool | None = False,
+        hidden_dropout: float | None = 0.0,
+        attention_dropout: float | None = 0.0,
+        bos_token_id: int | None = 0,
+        eos_token_id: int | None = 0,
+        pad_token_id: int | None = None,
         **kwargs,
     ):
         self.vocab_size = vocab_size
@@ -148,12 +148,11 @@ class StableLmConfig(PreTrainedConfig):
         self.rope_parameters = rope_parameters
         kwargs.setdefault("partial_rotary_factor", 0.25)  # assign default for BC
 
-        super().__init__(
-            bos_token_id=bos_token_id,
-            eos_token_id=eos_token_id,
-            tie_word_embeddings=tie_word_embeddings,
-            **kwargs,
-        )
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.tie_word_embeddings = tie_word_embeddings
+        super().__init__(**kwargs)
 
 
 __all__ = ["StableLmConfig"]

transformers/models/stablelm/modeling_stablelm.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 EleutherAI and the HuggingFace Inc. team. All rights reserved.
 #
 # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
@@ -45,7 +44,7 @@ from ...modeling_rope_utils import (
 from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
 from ...processing_utils import Unpack
 from ...utils import TransformersKwargs, auto_docstring, can_return_tuple, is_torch_flex_attn_available, logging
-from ...utils.generic import maybe_autocast
+from ...utils.generic import is_flash_attention_requested, maybe_autocast
 from .configuration_stablelm import StableLmConfig
 
 
@@ -81,9 +80,9 @@ class StableLmRotaryEmbedding(nn.Module):
     @staticmethod
     # Ignore copy
     def compute_default_rope_parameters(
-        config: Optional[StableLmConfig] = None,
+        config: StableLmConfig | 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
@@ -136,7 +135,7 @@ def rotate_half(x):
 
 
 # Copied from transformers.models.llama.modeling_llama.apply_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:
@@ -144,8 +143,6 @@ 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
@@ -214,7 +211,7 @@ 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],
@@ -238,7 +235,7 @@ def eager_attention_forward(
 class StableLmAttention(nn.Module):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
-    def __init__(self, config: StableLmConfig, layer_idx: Optional[int] = None):
+    def __init__(self, config: StableLmConfig, layer_idx: int | None = None):
         super().__init__()
         self.config = config
         self.layer_idx = layer_idx
@@ -281,15 +278,15 @@ class StableLmAttention(nn.Module):
     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,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
         output_attentions: bool = False,
         use_cache: bool = False,
-        cache_position: Optional[torch.LongTensor] = None,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
+        cache_position: torch.LongTensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
         **kwargs,
-    ) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
+    ) -> tuple[torch.Tensor, torch.Tensor | None, tuple[torch.Tensor] | None]:
         bsz, q_len, _ = hidden_states.size()
 
         query_states = self.q_proj(hidden_states)
@@ -330,9 +327,9 @@ class StableLmAttention(nn.Module):
             }
             key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
 
-        attention_interface: Callable = eager_attention_forward
-        if self.config._attn_implementation != "eager":
-            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
 
         attn_output, attn_weights = attention_interface(
             self,
@@ -368,14 +365,14 @@ class StableLmDecoderLayer(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,
-        output_attentions: Optional[bool] = False,
-        use_cache: Optional[bool] = False,
-        cache_position: Optional[torch.LongTensor] = None,
-        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,
-    ) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        output_attentions: bool | None = False,
+        use_cache: bool | None = False,
+        cache_position: torch.LongTensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+    ) -> tuple[torch.FloatTensor, tuple[torch.FloatTensor, torch.FloatTensor] | None]:
         """
         Args:
             hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
@@ -484,15 +481,15 @@ class StableLmModel(StableLmPreTrainedModel):
     @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,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: 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,
+        output_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs,
     ) -> BaseModelOutputWithPast:
         output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
@@ -578,7 +575,7 @@ class StableLmModel(StableLmPreTrainedModel):
         past_key_values: Cache,
         output_attentions: bool = False,
     ):
-        if self.config._attn_implementation == "flash_attention_2":
+        if is_flash_attention_requested(self.config):
             if attention_mask is not None and (attention_mask == 0.0).any():
                 return attention_mask
             return None
@@ -714,17 +711,17 @@ class StableLmForCausalLM(StableLmPreTrainedModel, GenerationMixin):
     # Ignore copy
     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_attentions: Optional[bool] = None,
-        output_hidden_states: 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_attentions: bool | None = None,
+        output_hidden_states: bool | None = None,
+        cache_position: torch.LongTensor | None = None,
+        logits_to_keep: int | torch.Tensor = 0,
         **kwargs,
     ) -> CausalLMOutputWithPast:
         r"""

transformers/models/starcoder2/configuration_starcoder2.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 BigCode and the HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
@@ -14,8 +13,6 @@
 # limitations under the License.
 """Starcoder2 model configuration"""
 
-from typing import Optional
-
 from ...configuration_utils import PreTrainedConfig
 from ...modeling_rope_utils import RopeParameters
 from ...utils import logging
@@ -70,6 +67,8 @@ class Starcoder2Config(PreTrainedConfig):
             The id of the "beginning-of-sequence" token.
         eos_token_id (`int`, *optional*, defaults to 50256):
             The id of the "end-of-sequence" token.
+        pad_token_id (`int`, *optional*):
+            Padding token id.
         rope_parameters (`RopeParameters`, *optional*):
             Dictionary containing the configuration parameters for the RoPE embeddings. The dictionary should contain
             a value for `rope_theta` and optionally parameters used for scaling in case you want to use RoPE
@@ -84,6 +83,8 @@ class Starcoder2Config(PreTrainedConfig):
             Embedding dropout.
         use_bias (`bool`, *optional*, defaults to `True`):
             Whether to use bias term on linear layers of the model.
+        tie_word_embeddings (`bool`, *optional*, defaults to `True`):
+            Whether to tie weight embeddings
 
 
     ```python
@@ -118,25 +119,27 @@ class Starcoder2Config(PreTrainedConfig):
 
     def __init__(
         self,
-        vocab_size: Optional[int] = 49152,
-        hidden_size: Optional[int] = 3072,
-        intermediate_size: Optional[int] = 12288,
-        num_hidden_layers: Optional[int] = 30,
-        num_attention_heads: Optional[int] = 24,
-        num_key_value_heads: Optional[int] = 2,
-        hidden_act: Optional[str] = "gelu_pytorch_tanh",
-        max_position_embeddings: Optional[int] = 4096,
-        initializer_range: Optional[float] = 0.018042,
-        norm_epsilon: Optional[int] = 1e-5,
-        use_cache: Optional[bool] = True,
-        bos_token_id: Optional[int] = 50256,
-        eos_token_id: Optional[int] = 50256,
-        rope_parameters: Optional[RopeParameters | dict[str, RopeParameters]] = None,
-        sliding_window: Optional[int] = None,
-        attention_dropout: Optional[float] = 0.0,
-        residual_dropout: Optional[float] = 0.0,
-        embedding_dropout: Optional[float] = 0.0,
-        use_bias: Optional[bool] = True,
+        vocab_size: int | None = 49152,
+        hidden_size: int | None = 3072,
+        intermediate_size: int | None = 12288,
+        num_hidden_layers: int | None = 30,
+        num_attention_heads: int | None = 24,
+        num_key_value_heads: int | None = 2,
+        hidden_act: str | None = "gelu_pytorch_tanh",
+        max_position_embeddings: int | None = 4096,
+        initializer_range: float | None = 0.018042,
+        norm_epsilon: int | None = 1e-5,
+        use_cache: bool | None = True,
+        bos_token_id: int | None = 50256,
+        eos_token_id: int | None = 50256,
+        pad_token_id: int | None = None,
+        rope_parameters: RopeParameters | dict[str, RopeParameters] | None = None,
+        sliding_window: int | None = None,
+        attention_dropout: float | None = 0.0,
+        residual_dropout: float | None = 0.0,
+        embedding_dropout: float | None = 0.0,
+        use_bias: bool | None = True,
+        tie_word_embeddings: bool | None = True,
         **kwargs,
     ):
         self.vocab_size = vocab_size
@@ -157,11 +160,11 @@ class Starcoder2Config(PreTrainedConfig):
         self.embedding_dropout = embedding_dropout
         self.rope_parameters = rope_parameters
 
-        super().__init__(
-            bos_token_id=bos_token_id,
-            eos_token_id=eos_token_id,
-            **kwargs,
-        )
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.tie_word_embeddings = tie_word_embeddings
+        super().__init__(**kwargs)
 
 
 __all__ = ["Starcoder2Config"]

transformers/models/starcoder2/modeling_starcoder2.py

@@ -4,7 +4,6 @@
 #             the file from the modular. If any change should be done, please apply the change to the
 #                          modular_starcoder2.py file directly. One of our CI enforces this.
 #                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
-# coding=utf-8
 # Copyright 2024 BigCode and the HuggingFace Inc. team. All rights reserved.
 #
 # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
@@ -25,7 +24,7 @@
 # limitations under the License.
 
 from collections.abc import Callable
-from typing import Optional, Union
+from typing import Optional
 
 import torch
 from torch import nn
@@ -61,7 +60,7 @@ class Starcoder2MLP(nn.Module):
         self.act = ACT2FN[config.hidden_act]
         self.residual_dropout = config.residual_dropout
 
-    def forward(self, hidden_states: Optional[tuple[torch.FloatTensor]]) -> torch.FloatTensor:
+    def forward(self, hidden_states: tuple[torch.FloatTensor] | None) -> torch.FloatTensor:
         hidden_states = self.c_fc(hidden_states)
         hidden_states = self.act(hidden_states)
         hidden_states = self.c_proj(hidden_states)
@@ -77,7 +76,7 @@ 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:
@@ -85,8 +84,6 @@ 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
@@ -121,7 +118,7 @@ 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],
@@ -146,7 +143,7 @@ def eager_attention_forward(
 class Starcoder2Attention(nn.Module):
     """Multi-headed attention from 'Attention Is All You Need' paper"""
 
-    def __init__(self, config: Starcoder2Config, layer_idx: Optional[int] = None):
+    def __init__(self, config: Starcoder2Config, layer_idx: int | None = None):
         super().__init__()
         self.config = config
         self.layer_idx = layer_idx
@@ -165,11 +162,11 @@ class Starcoder2Attention(nn.Module):
         self,
         hidden_states: torch.Tensor,
         position_embeddings: tuple[torch.Tensor, torch.Tensor],
-        attention_mask: Optional[torch.Tensor],
-        past_key_values: Optional[Cache] = None,
-        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: torch.Tensor | None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs: Unpack[FlashAttentionKwargs],
-    ) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
+    ) -> tuple[torch.Tensor, torch.Tensor | None, tuple[torch.Tensor] | None]:
         input_shape = hidden_states.shape[:-1]
         hidden_shape = (*input_shape, -1, self.head_dim)
 
@@ -185,9 +182,9 @@ class Starcoder2Attention(nn.Module):
             cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
             key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
 
-        attention_interface: Callable = eager_attention_forward
-        if self.config._attn_implementation != "eager":
-            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
 
         attn_output, attn_weights = attention_interface(
             self,
@@ -222,12 +219,12 @@ class Starcoder2DecoderLayer(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
@@ -293,9 +290,9 @@ class Starcoder2RotaryEmbedding(nn.Module):
 
     @staticmethod
     def compute_default_rope_parameters(
-        config: Optional[Starcoder2Config] = None,
+        config: Starcoder2Config | 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
@@ -359,15 +356,15 @@ class Starcoder2Model(Starcoder2PreTrainedModel):
     @check_model_inputs
     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],
-    ) -> BaseModelOutputWithPast:
+    ) -> tuple | BaseModelOutputWithPast:
         if (input_ids is None) ^ (inputs_embeds is not None):
             raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
 
@@ -425,7 +422,7 @@ class Starcoder2Model(Starcoder2PreTrainedModel):
 @auto_docstring
 class Starcoder2ForCausalLM(Starcoder2PreTrainedModel, GenerationMixin):
     _tied_weights_keys = {"lm_head.weight": "model.embed_tokens.weight"}
-    _tp_plan = {"lm_head": "colwise_rep"}
+    _tp_plan = {"lm_head": "colwise_gather_output"}
     _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
 
     def __init__(self, config):
@@ -441,15 +438,15 @@ class Starcoder2ForCausalLM(Starcoder2PreTrainedModel, 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/starcoder2/modular_starcoder2.py

@@ -1,4 +1,3 @@
-# coding=utf-8
 # Copyright 2024 BigCode and the HuggingFace Inc. team. All rights reserved.
 #
 # This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
@@ -20,7 +19,6 @@
 """PyTorch Starcoder2 model."""
 
 from collections.abc import Callable
-from typing import Optional
 
 import torch
 from torch import nn
@@ -60,7 +58,7 @@ class Starcoder2MLP(nn.Module):
         self.act = ACT2FN[config.hidden_act]
         self.residual_dropout = config.residual_dropout
 
-    def forward(self, hidden_states: Optional[tuple[torch.FloatTensor]]) -> torch.FloatTensor:
+    def forward(self, hidden_states: tuple[torch.FloatTensor] | None) -> torch.FloatTensor:
         hidden_states = self.c_fc(hidden_states)
         hidden_states = self.act(hidden_states)
         hidden_states = self.c_proj(hidden_states)
@@ -69,7 +67,7 @@ class Starcoder2MLP(nn.Module):
 
 
 class Starcoder2Attention(MistralAttention):
-    def __init__(self, config: Starcoder2Config, layer_idx: Optional[int] = None):
+    def __init__(self, config: Starcoder2Config, layer_idx: int | None = None):
         super().__init__(config=config, layer_idx=layer_idx)
         self.residual_dropout = config.residual_dropout
         self.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.use_bias)
@@ -81,11 +79,11 @@ class Starcoder2Attention(MistralAttention):
         self,
         hidden_states: torch.Tensor,
         position_embeddings: tuple[torch.Tensor, torch.Tensor],
-        attention_mask: Optional[torch.Tensor],
-        past_key_values: Optional[Cache] = None,
-        cache_position: Optional[torch.LongTensor] = None,
+        attention_mask: torch.Tensor | None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
         **kwargs: Unpack[FlashAttentionKwargs],
-    ) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
+    ) -> tuple[torch.Tensor, torch.Tensor | None, tuple[torch.Tensor] | None]:
         input_shape = hidden_states.shape[:-1]
         hidden_shape = (*input_shape, -1, self.head_dim)
 
@@ -101,9 +99,9 @@ class Starcoder2Attention(MistralAttention):
             cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
             key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
 
-        attention_interface: Callable = eager_attention_forward
-        if self.config._attn_implementation != "eager":
-            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
 
         attn_output, attn_weights = attention_interface(
             self,
@@ -147,15 +145,15 @@ class Starcoder2Model(MistralModel):
     @check_model_inputs
     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],
-    ) -> BaseModelOutputWithPast:
+    ) -> tuple | BaseModelOutputWithPast:
         if (input_ids is None) ^ (inputs_embeds is not None):
             raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
 

transformers/models/superglue/configuration_superglue.py

@@ -11,7 +11,7 @@
 # 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 TYPE_CHECKING, Optional
+from typing import TYPE_CHECKING
 
 from ...configuration_utils import PreTrainedConfig
 from ...utils import logging
@@ -51,6 +51,9 @@ class SuperGlueConfig(PreTrainedConfig):
             The matching threshold.
         initializer_range (`float`, *optional*, defaults to 0.02):
             The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        is_decoder (`bool`, *optional*, defaults to `False`):
+            Whether to only use the decoder in an encoder-decoder architecture, otherwise it has no effect on
+            decoder-only or encoder-only architectures.
 
     Examples:
         ```python
@@ -74,12 +77,13 @@ class SuperGlueConfig(PreTrainedConfig):
         self,
         keypoint_detector_config: "SuperPointConfig" = None,
         hidden_size: int = 256,
-        keypoint_encoder_sizes: Optional[list[int]] = None,
-        gnn_layers_types: Optional[list[str]] = None,
+        keypoint_encoder_sizes: list[int] | None = None,
+        gnn_layers_types: list[str] | None = None,
         num_attention_heads: int = 4,
         sinkhorn_iterations: int = 100,
         matching_threshold: float = 0.0,
         initializer_range: float = 0.02,
+        is_decoder=False,
         **kwargs,
     ):
         self.gnn_layers_types = gnn_layers_types if gnn_layers_types is not None else ["self", "cross"] * 9