Update modeling_gpt2.py

Updating based on transformers==4.52.4

modeling_gpt2.py

@@ -15,119 +15,23 @@
 # limitations under the License.
 """PyTorch OpenAI GPT-2 model."""
 
-import math
-import os
-import warnings
 from dataclasses import dataclass
-from typing import Optional, Tuple, Union
+from typing import Callable, Optional, Tuple, Union
 
 import torch
-import torch.utils.checkpoint
-from packaging import version
 from torch import nn
-from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
 
-from transformers.activations import ACT2FN
-from transformers.generation import GenerationMixin
-from transformers.modeling_attn_mask_utils import (
-    _prepare_4d_attention_mask_for_sdpa,
-    _prepare_4d_causal_attention_mask_for_sdpa,
-)
-from transformers.modeling_outputs import (
-    BaseModelOutputWithPastAndCrossAttentions,
-    CausalLMOutputWithCrossAttentions,
-    QuestionAnsweringModelOutput,
-    SequenceClassifierOutputWithPast,
-    TokenClassifierOutput,
-)
-from transformers.modeling_utils import PreTrainedModel, SequenceSummary
-from transformers.pytorch_utils import (
-    Conv1D,
-    find_pruneable_heads_and_indices,
-    prune_conv1d_layer,
-)
+from transformers.cache_utils import Cache, DynamicCache, EncoderDecoderCache, StaticCache
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
+from transformers.pytorch_utils import Conv1D, find_pruneable_heads_and_indices, prune_conv1d_layer
 from transformers.utils import (
-    ModelOutput,
-    add_code_sample_docstrings,
-    add_start_docstrings,
-    add_start_docstrings_to_model_forward,
-    get_torch_version,
-    is_flash_attn_2_available,
-    is_flash_attn_greater_or_equal_2_10,
     logging,
-    replace_return_docstrings,
 )
-from transformers.utils.model_parallel_utils import assert_device_map, get_device_map
-from .configuration_gpt2 import GPT2Config
-
-
-if is_flash_attn_2_available():
-    from transformers.modeling_flash_attention_utils import _flash_attention_forward
-
+from transformers.utils.deprecation import deprecate_kwarg
+from transformers.models.gpt2.modeling_gpt2 import load_tf_weights_in_gpt2, eager_attention_forward, GPT2Block, GPT2MLP, GPT2SequenceSummary,GPT2PreTrainedModel,GPT2DoubleHeadsModelOuptut,GPT2DoubleHeadsModel, GPT2Model,GPT2LMHeadModel, GPT2DoubleHeadsModel, GPT2ForSequenceClassification,GPT2ForTokenClassification,GPT2ForQuestionAnswering
 
 logger = logging.get_logger(__name__)
 
-_CHECKPOINT_FOR_DOC = "openai-community/gpt2"
-_CONFIG_FOR_DOC = "GPT2Config"
-
-
-def load_tf_weights_in_gpt2(model, config, gpt2_checkpoint_path):
-    """Load tf checkpoints in a pytorch model"""
-    try:
-        import re
-
-        import tensorflow as tf
-    except ImportError:
-        logger.error(
-            "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
-            "https://www.tensorflow.org/install/ for installation instructions."
-        )
-        raise
-    tf_path = os.path.abspath(gpt2_checkpoint_path)
-    logger.info(f"Converting TensorFlow checkpoint from {tf_path}")
-    # Load weights from TF model
-    init_vars = tf.train.list_variables(tf_path)
-    names = []
-    arrays = []
-    for name, shape in init_vars:
-        logger.info(f"Loading TF weight {name} with shape {shape}")
-        array = tf.train.load_variable(tf_path, name)
-        names.append(name)
-        arrays.append(array.squeeze())
-
-    for name, array in zip(names, arrays):
-        name = name[6:]  # skip "model/"
-        name = name.split("/")
-        pointer = model
-        for m_name in name:
-            if re.fullmatch(r"[A-Za-z]+\d+", m_name):
-                scope_names = re.split(r"(\d+)", m_name)
-            else:
-                scope_names = [m_name]
-            if scope_names[0] == "w" or scope_names[0] == "g":
-                pointer = getattr(pointer, "weight")
-            elif scope_names[0] == "b":
-                pointer = getattr(pointer, "bias")
-            elif scope_names[0] == "wpe" or scope_names[0] == "wte":
-                pointer = getattr(pointer, scope_names[0])
-                pointer = getattr(pointer, "weight")
-            else:
-                pointer = getattr(pointer, scope_names[0])
-            if len(scope_names) >= 2:
-                num = int(scope_names[1])
-                pointer = pointer[num]
-        try:
-            if pointer.shape != array.shape:
-                raise ValueError(
-                    f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched"
-                )
-        except ValueError as e:
-            e.args += (pointer.shape, array.shape)
-            raise
-        logger.info(f"Initialize PyTorch weight {name}")
-        pointer.data = torch.from_numpy(array)
-    return model
-
 
 class GPT2Attention(nn.Module):
     def __init__(self, config, is_cross_attention=False, layer_idx=None):
@@ -136,9 +40,9 @@ class GPT2Attention(nn.Module):
         max_positions = config.max_position_embeddings
         self.register_buffer(
             "bias",
-            torch.tril(
-                torch.ones((max_positions, max_positions), dtype=torch.bool)
-            ).view(1, 1, max_positions, max_positions),
+            torch.tril(torch.ones((max_positions, max_positions), dtype=torch.bool)).view(
+                1, 1, max_positions, max_positions
+            ),
             persistent=False,
         )
         self.register_buffer("masked_bias", torch.tensor(-1e4), persistent=False)
@@ -177,88 +81,25 @@ class GPT2Attention(nn.Module):
     def prune_heads(self, heads):
         if len(heads) == 0:
             return
-        heads, index = find_pruneable_heads_and_indices(
-            heads, self.num_heads, self.head_dim, self.pruned_heads
-        )
-        index_attn = torch.cat(
-            [index, index + self.split_size, index + (2 * self.split_size)]
-        )
+        heads, index = find_pruneable_heads_and_indices(heads, self.num_heads, self.head_dim, self.pruned_heads)
+        index_attn = torch.cat([index, index + self.split_size, index + (2 * self.split_size)])
 
         # Prune conv1d layers
         self.c_attn = prune_conv1d_layer(self.c_attn, index_attn, dim=1)
         self.c_proj = prune_conv1d_layer(self.c_proj, index, dim=0)
 
         # Update hyper params
-        self.split_size = (self.split_size // self.num_heads) * (
-            self.num_heads - len(heads)
-        )
+        self.split_size = (self.split_size // self.num_heads) * (self.num_heads - len(heads))
         self.num_heads = self.num_heads - len(heads)
         self.pruned_heads = self.pruned_heads.union(heads)
 
-    def _attn(self, query, key, value, attention_mask=None, head_mask=None):
-        attn_weights = torch.matmul(query, key.transpose(-1, -2))
-
-        if self.scale_attn_weights:
-            attn_weights = attn_weights / torch.full(
-                [],
-                value.size(-1) ** 0.5,
-                dtype=attn_weights.dtype,
-                device=attn_weights.device,
-            )
-
-        # Layer-wise attention scaling
-        if self.scale_attn_by_inverse_layer_idx:
-            attn_weights = attn_weights / float(self.layer_idx + 1)
-
-        if not self.is_cross_attention:
-            # if only "normal" attention layer implements causal mask
-            query_length, key_length = query.size(-2), key.size(-2)
-            causal_mask = self.bias[
-                :, :, key_length - query_length : key_length, :key_length
-            ]
-            mask_value = torch.finfo(attn_weights.dtype).min
-            # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
-            # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
-            mask_value = torch.full(
-                [], mask_value, dtype=attn_weights.dtype, device=attn_weights.device
-            )
-            attn_weights = torch.where(
-                causal_mask, attn_weights.to(attn_weights.dtype), mask_value
-            )
-
-        if attention_mask is not None:
-            # Apply the attention mask
-            attn_weights = attn_weights + attention_mask
-
-        attn_weights = nn.functional.softmax(attn_weights, dim=-1)
-
-        # Downcast (if necessary) back to V's dtype (if in mixed-precision) -- No-Op otherwise
-        attn_weights = attn_weights.type(value.dtype)
-        attn_weights = self.attn_dropout(attn_weights)
-
-        # Mask heads if we want to
-        if head_mask is not None:
-            attn_weights = attn_weights * head_mask
-
-        attn_output = torch.matmul(attn_weights, value)
-
-        return attn_output, attn_weights
-
-    def _upcast_and_reordered_attn(
-        self, query, key, value, attention_mask=None, head_mask=None
-    ):
+    def _upcast_and_reordered_attn(self, query, key, value, attention_mask=None, head_mask=None):
         # Use `torch.baddbmm` (a bit more efficient w/ alpha param for scaling -- from Megatron-LM)
         bsz, num_heads, q_seq_len, dk = query.size()
         _, _, k_seq_len, _ = key.size()
 
         # Preallocate attn_weights for `baddbmm`
-        attn_weights = torch.empty(
-            bsz * num_heads,
-            q_seq_len,
-            k_seq_len,
-            dtype=torch.float32,
-            device=query.device,
-        )
+        attn_weights = torch.empty(bsz * num_heads, q_seq_len, k_seq_len, dtype=torch.float32, device=query.device)
 
         # Compute Scale Factor
         scale_factor = 1.0
@@ -270,26 +111,18 @@ class GPT2Attention(nn.Module):
 
         # Upcast (turn off autocast) and reorder (Scale K by 1 / root(dk))
         with torch.amp.autocast(query.device.type, enabled=False):
-            q, k = query.reshape(-1, q_seq_len, dk), key.transpose(-1, -2).reshape(
-                -1, dk, k_seq_len
-            )
-            attn_weights = torch.baddbmm(
-                attn_weights, q.float(), k.float(), beta=0, alpha=scale_factor
-            )
+            q, k = query.reshape(-1, q_seq_len, dk), key.transpose(-1, -2).reshape(-1, dk, k_seq_len)
+            attn_weights = torch.baddbmm(attn_weights, q.float(), k.float(), beta=0, alpha=scale_factor)
             attn_weights = attn_weights.reshape(bsz, num_heads, q_seq_len, k_seq_len)
 
         if not self.is_cross_attention:
             # if only "normal" attention layer implements causal mask
             query_length, key_length = query.size(-2), key.size(-2)
-            causal_mask = self.bias[
-                :, :, key_length - query_length : key_length, :key_length
-            ]
+            causal_mask = self.bias[:, :, key_length - query_length : key_length, :key_length]
             mask_value = torch.finfo(attn_weights.dtype).min
             # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
             # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
-            mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype).to(
-                attn_weights.device
-            )
+            mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype, device=attn_weights.device)
             attn_weights = torch.where(causal_mask, attn_weights, mask_value)
 
         if attention_mask is not None:
@@ -300,9 +133,7 @@ class GPT2Attention(nn.Module):
 
         # Downcast (if necessary) back to V's dtype (if in mixed-precision) -- No-Op if otherwise
         if attn_weights.dtype != torch.float32:
-            raise RuntimeError(
-                "Error with upcasting, attn_weights does not have dtype torch.float32"
-            )
+            raise RuntimeError("Error with upcasting, attn_weights does not have dtype torch.float32")
         attn_weights = attn_weights.type(value.dtype)
         attn_weights = self.attn_dropout(attn_weights)
 
@@ -311,1744 +142,102 @@ class GPT2Attention(nn.Module):
             attn_weights = attn_weights * head_mask
 
         attn_output = torch.matmul(attn_weights, value)
+        attn_output = attn_output.transpose(1, 2)
 
         return attn_output, attn_weights
 
-    def _split_heads(self, tensor, num_heads, attn_head_size):
-        """
-        Splits hidden_size dim into attn_head_size and num_heads
-        """
-        new_shape = tensor.size()[:-1] + (num_heads, attn_head_size)
-        tensor = tensor.view(new_shape)
-        return tensor.permute(0, 2, 1, 3)  # (batch, head, seq_length, head_features)
-
-    def _merge_heads(self, tensor, num_heads, attn_head_size):
-        """
-        Merges attn_head_size dim and num_attn_heads dim into hidden_size
-        """
-        tensor = tensor.permute(0, 2, 1, 3).contiguous()
-        new_shape = tensor.size()[:-2] + (num_heads * attn_head_size,)
-        return tensor.view(new_shape)
-
+    @deprecate_kwarg("layer_past", new_name="past_key_value", version="4.53.0", raise_if_both_names=True)
     def forward(
         self,
         hidden_states: Optional[Tuple[torch.FloatTensor]],
-        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        past_key_value: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
         attention_mask: Optional[torch.FloatTensor] = None,
         head_mask: Optional[torch.FloatTensor] = None,
         encoder_hidden_states: Optional[torch.Tensor] = None,
         encoder_attention_mask: Optional[torch.FloatTensor] = None,
-        use_cache: Optional[bool] = False,
         output_attentions: Optional[bool] = False,
+        **kwargs,
     ) -> Tuple[Union[torch.Tensor, Tuple[torch.Tensor]], ...]:
-        if encoder_hidden_states is not None:
+        is_cross_attention = encoder_hidden_states is not None
+        if is_cross_attention:
             if not hasattr(self, "q_attn"):
                 raise ValueError(
                     "If class is used as cross attention, the weights `q_attn` have to be defined. "
                     "Please make sure to instantiate class with `GPT2Attention(..., is_cross_attention=True)`."
                 )
 
-            query = self.q_attn(hidden_states)
-            key, value = self.c_attn(encoder_hidden_states).split(
-                self.split_size, dim=2
-            )
+            query_states = self.q_attn(hidden_states)
+            key_states, value_states = self.c_attn(encoder_hidden_states).split(self.split_size, dim=2)
             attention_mask = encoder_attention_mask
         else:
-            query, key, value = self.c_attn(hidden_states).split(self.split_size, dim=2)
-
-        query = self._split_heads(query, self.num_heads, self.head_dim)
-        key = self._split_heads(key, self.num_heads, self.head_dim)
-        value = self._split_heads(value, self.num_heads, self.head_dim)
+            query_states, key_states, value_states = self.c_attn(hidden_states).split(self.split_size, dim=2)
 
-        if layer_past is not None:
-            past_key, past_value = layer_past
-            key = torch.cat((past_key, key), dim=-2)
-            value = torch.cat((past_value, value), dim=-2)
+        shape_q = (query_states.shape[0],query_states.shape[1], -1, self.head_dim)
+        shape_kv = (query_states.shape[0],query_states.shape[1], -1, self.head_dim)
 
-        if use_cache is True:
-            present = (key, value)
-        else:
-            present = None
+        query_states = query_states.view(shape_q).transpose(1, 2)
+        key_states = key_states.view(shape_kv).transpose(1, 2)
+        value_states = value_states.view(shape_kv).transpose(1, 2)
 
-        if self.reorder_and_upcast_attn:
-            attn_output, attn_weights = self._upcast_and_reordered_attn(
-                query, key, value, attention_mask, head_mask
-            )
-        else:
-            attn_output, attn_weights = self._attn(
-                query, key, value, attention_mask, head_mask
+        if past_key_value is not None:
+            if isinstance(past_key_value, EncoderDecoderCache):
+                if is_cross_attention:
+                    past_key_value = past_key_value.cross_attention_cache
+                else:
+                    past_key_value = past_key_value.self_attention_cache
+            cache_kwargs = {"cache_position": cache_position}
+            key_states, value_states = past_key_value.update(
+                key_states, value_states, self.layer_idx, cache_kwargs=cache_kwargs
             )
 
-        attn_output = self._merge_heads(attn_output, self.num_heads, self.head_dim)
-        attn_output = self.c_proj(attn_output)
-        attn_output = self.resid_dropout(attn_output)
-
-        outputs = (attn_output, present)
-        if output_attentions:
-            outputs += (attn_weights,)
-
-        return outputs  # a, present, (attentions)
-
-
-class GPT2FlashAttention2(GPT2Attention):
-    """
-    GPT2 flash attention module. This module inherits from `GPT2Attention` as the weights of the module stays
-    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
-    flash attention and deal with padding tokens in case the input contains any of them.
-    """
+        is_causal = attention_mask is None and query_states.shape[-2] > 1 and not is_cross_attention
 
-    # Copied from transformers.models.llama.modeling_llama.LlamaFlashAttention2.__init__
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-
-        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
-        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
-        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
-        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
-
-    def forward(
-        self,
-        hidden_states: Optional[Tuple[torch.FloatTensor]],
-        layer_past: Optional[Tuple[torch.Tensor]] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        head_mask: Optional[torch.FloatTensor] = None,
-        encoder_hidden_states: Optional[torch.Tensor] = None,
-        encoder_attention_mask: Optional[torch.FloatTensor] = None,
-        use_cache: Optional[bool] = False,
-        output_attentions: Optional[bool] = False,
-    ) -> Tuple[Union[torch.Tensor, Tuple[torch.Tensor]], ...]:
-        bsz, _, _ = hidden_states.size()
-        if encoder_hidden_states is not None:
-            if not hasattr(self, "q_attn"):
-                raise ValueError(
-                    "If class is used as cross attention, the weights `q_attn` have to be defined. "
-                    "Please make sure to instantiate class with `GPT2Attention(..., is_cross_attention=True)`."
+        using_eager = self.config._attn_implementation == "eager"
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            if self.config._attn_implementation == "sdpa" and (output_attentions or head_mask is not None):
+                using_eager = True
+                logger.warning_once(
+                    "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
+                    'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
                 )
-
-            query = self.q_attn(hidden_states)
-            key, value = self.c_attn(encoder_hidden_states).split(
-                self.split_size, dim=2
-            )
-            attention_mask = encoder_attention_mask
-        else:
-            query, key, value = self.c_attn(hidden_states).split(self.split_size, dim=2)
-
-        query = self._split_heads(query, self.num_heads, self.head_dim)
-        key = self._split_heads(key, self.num_heads, self.head_dim)
-        value = self._split_heads(value, self.num_heads, self.head_dim)
-
-        if layer_past is not None:
-            past_key = layer_past[0]
-            past_value = layer_past[1]
-            key = torch.cat((past_key, key), dim=-2)
-            value = torch.cat((past_value, value), dim=-2)
-
-        present = None
-        if use_cache is True:
-            present = (key, value)
-
-        query_length = query.shape[2]
-        tgt_len = key.shape[2]
-
-        # Flash attention requires the input to have the shape
-        # batch_size x seq_length x head_dim x hidden_dim
-        query = query.transpose(1, 2).view(
-            bsz, query_length, self.num_heads, self.head_dim
-        )
-        key = key.transpose(1, 2).view(bsz, tgt_len, self.num_heads, self.head_dim)
-        value = value.transpose(1, 2).view(bsz, tgt_len, self.num_heads, self.head_dim)
-
-        attn_dropout = self.attn_dropout.p if self.training else 0.0
-
-        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
-        # therefore the input hidden states gets silently casted in float32. Hence, we need
-        # cast them back in the correct dtype just to be sure everything works as expected.
-        # This might slowdown training & inference so it is recommended to not cast the LayerNorms
-        # in fp32. (LlamaRMSNorm handles it correctly)
-
-        if query.dtype == torch.float32:
-            if torch.is_autocast_enabled():
-                target_dtype = torch.get_autocast_gpu_dtype()
-            # Handle the case where the model is quantized
-            elif hasattr(self.config, "_pre_quantization_dtype"):
-                target_dtype = self.config._pre_quantization_dtype
             else:
-                target_dtype = self.c_proj.weight.dtype
-
-            logger.warning_once(
-                f"The input hidden states seems to be silently casted in float32, this might be related to"
-                f" the fact you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
-                f" {target_dtype}."
-            )
-
-            query = query.to(target_dtype)
-            key = key.to(target_dtype)
-            value = value.to(target_dtype)
-
-        attn_output = _flash_attention_forward(
-            query,
-            key,
-            value,
-            attention_mask,
-            query_length,
-            dropout=attn_dropout,
-            is_causal=self.is_causal,
-            use_top_left_mask=self._flash_attn_uses_top_left_mask,
-        )
-
-        attn_weights_reshaped = attn_output.reshape(
-            bsz, query_length, self.num_heads * self.head_dim
-        )
-        attn_output = self.c_proj(attn_weights_reshaped)
-        attn_output = self.resid_dropout(attn_output)
+                # Attention functions are consistent with previous equivalent attention classes, however they do not support some options
+                # (e.g. layer scaling, head mask) that eager supports. These implementations are thus equivalent to previous code, but
+                # not necessarily to eager (if mentioned options are provided).
+                attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
 
-        outputs = (attn_output, present)
-        if output_attentions:
-            outputs += (attn_weights_reshaped,)
-
-        return outputs
-
-
-class GPT2SdpaAttention(GPT2Attention):
-    """
-    GPT2 attention module using torch.nn.functional.scaled_dot_product_attention. This module inherits from
-    `GPT2Attention` as the weights of the module stays untouched. The only changes are on the forward pass
-    to adapt to the SDPA API.
-    """
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-
-        # Idea adapted from transformers.models.bert.modeling_bert.BertSdpaSelfAttention.__init__
-        # SDPA with memory-efficient backend is broken in torch==2.1.2 when using non-contiguous inputs and a custom
-        # attn_mask, so we need to call `.contiguous()`. This was fixed in torch==2.2.0.
-        # Reference: https://github.com/pytorch/pytorch/issues/112577
-        self.require_contiguous_qkv = version.parse(
-            get_torch_version()
-        ) < version.parse("2.2.0")
-
-    def forward(
-        self,
-        hidden_states: Optional[Tuple[torch.FloatTensor]],
-        layer_past: Optional[Tuple[torch.Tensor]] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        head_mask: Optional[torch.FloatTensor] = None,
-        encoder_hidden_states: Optional[torch.Tensor] = None,
-        encoder_attention_mask: Optional[torch.FloatTensor] = None,
-        use_cache: Optional[bool] = False,
-        output_attentions: Optional[bool] = False,
-    ) -> Tuple[Union[torch.Tensor, Tuple[torch.Tensor]], ...]:
-        if output_attentions or head_mask is not None:
-            logger.warning_once(
-                "`GPT2SdpaAttention` is used but `torch.nn.functional.scaled_dot_product_attention` does not support "
-                "`output_attentions=True` or `head_mask`. Falling back to the manual attention implementation, but "
-                "specifying the manual implementation will be required from Transformers version v5.0.0 onwards. "
-                'This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
+        if using_eager and self.reorder_and_upcast_attn:
+            attn_output, attn_weights = self._upcast_and_reordered_attn(
+                query_states, key_states, value_states, attention_mask, head_mask
             )
-            return super().forward(
-                hidden_states=hidden_states,
-                layer_past=layer_past,
-                attention_mask=attention_mask,
+        else:
+            attn_output, attn_weights = attention_interface(
+                self,
+                query_states,
+                key_states,
+                value_states,
+                attention_mask,
                 head_mask=head_mask,
-                encoder_hidden_states=encoder_hidden_states,
-                encoder_attention_mask=encoder_attention_mask,
-                use_cache=use_cache,
-                output_attentions=output_attentions,
-            )
-
-        bsz, q_len, _ = hidden_states.size()
-
-        # Initial attention projections
-        is_cross_attention = encoder_hidden_states is not None
-        if is_cross_attention:
-            if not hasattr(self, "q_attn"):
-                raise ValueError(
-                    "If class is used as cross attention, the weights `q_attn` have to be defined. "
-                    "Please make sure to instantiate class with `GPT2SdpaAttention(..., is_cross_attention=True)`."
-                )
-
-            query = self.q_attn(hidden_states)
-            key, value = self.c_attn(encoder_hidden_states).split(
-                self.split_size, dim=2
+                dropout=self.attn_dropout.p if self.training else 0.0,
+                is_causal=is_causal,
+                **kwargs,
             )
-            attention_mask = encoder_attention_mask
-        else:
-            query, key, value = self.c_attn(hidden_states).split(self.split_size, dim=2)
-
-        query = self._split_heads(query, self.num_heads, self.head_dim)
-        key = self._split_heads(key, self.num_heads, self.head_dim)
-        value = self._split_heads(value, self.num_heads, self.head_dim)
-
-        # Optional kv caching
-        if layer_past is not None:
-            past_key = layer_past[0]
-            past_value = layer_past[1]
-            key = torch.cat((past_key, key), dim=-2)
-            value = torch.cat((past_value, value), dim=-2)
-
-        present = None
-        if use_cache is True:
-            present = (key, value)
-
-        # Avoid torch==2.1.2 specific bug for the memory-efficient backend in SDPA
-        if (
-            self.require_contiguous_qkv
-            and query.device.type == "cuda"
-            and attention_mask is not None
-        ):
-            query = query.contiguous()
-            key = key.contiguous()
-            value = value.contiguous()
-
-        # We dispatch to SDPA's Flash Attention or Efficient kernels via this `is_causal` if statement instead of an inline conditional assignment
-        # in SDPA to support both torch.compile's dynamic shapes and full graph options. An inline conditional prevents dynamic shapes from compiling.
-        is_causal = (
-            True
-            if attention_mask is None and q_len > 1 and not is_cross_attention
-            else False
-        )
 
-        attn_output = torch.nn.functional.scaled_dot_product_attention(
-            query,
-            key,
-            value,
-            attn_mask=attention_mask,
-            dropout_p=self.attn_dropout.p if self.training else 0.0,
-            is_causal=is_causal,
-        )
-
-        # Reshape outputs
-        attn_output = attn_output.transpose(1, 2).contiguous()
-        attn_output = attn_output.view(bsz, q_len, self.embed_dim)
-
-        # Final projection
+        attn_output = attn_output.reshape(attn_output.shape[0],attn_output.shape[1], -1).contiguous()
         attn_output = self.c_proj(attn_output)
         attn_output = self.resid_dropout(attn_output)
 
-        return attn_output, present, None
-
-
-class GPT2MLP(nn.Module):
-    def __init__(self, intermediate_size, config):
-        super().__init__()
-        embed_dim = config.hidden_size
-        self.c_fc = Conv1D(intermediate_size, embed_dim)
-        self.c_proj = Conv1D(embed_dim, intermediate_size)
-        self.act = ACT2FN[config.activation_function]
-        self.dropout = nn.Dropout(config.resid_pdrop)
-
-    def forward(
-        self, hidden_states: Optional[Tuple[torch.FloatTensor]]
-    ) -> torch.FloatTensor:
-        hidden_states = self.c_fc(hidden_states)
-        hidden_states = self.act(hidden_states)
-        hidden_states = self.c_proj(hidden_states)
-        hidden_states = self.dropout(hidden_states)
-        return hidden_states
-
-
-GPT2_ATTENTION_CLASSES = {
-    "eager": GPT2Attention,
-    "flash_attention_2": GPT2FlashAttention2,
-    "sdpa": GPT2SdpaAttention,
-}
-
-
-class GPT2Block(nn.Module):
-    def __init__(self, config, layer_idx=None):
-        super().__init__()
-        hidden_size = config.hidden_size
-        inner_dim = config.n_inner if config.n_inner is not None else 4 * hidden_size
-        attention_class = GPT2_ATTENTION_CLASSES[config._attn_implementation]
-
-        self.ln_1 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
-        self.attn = attention_class(config=config, layer_idx=layer_idx)
-        self.ln_2 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
-
-        if config.add_cross_attention:
-            self.crossattention = attention_class(
-                config=config, is_cross_attention=True, layer_idx=layer_idx
-            )
-            self.ln_cross_attn = nn.LayerNorm(
-                hidden_size, eps=config.layer_norm_epsilon
-            )
-
-        self.mlp = GPT2MLP(inner_dim, config)
-
-    def forward(
-        self,
-        hidden_states: Optional[Tuple[torch.FloatTensor]],
-        layer_past: Optional[Tuple[torch.Tensor]] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        head_mask: Optional[torch.FloatTensor] = None,
-        encoder_hidden_states: Optional[torch.Tensor] = None,
-        encoder_attention_mask: Optional[torch.FloatTensor] = None,
-        use_cache: Optional[bool] = False,
-        output_attentions: Optional[bool] = False,
-    ) -> Union[
-        Tuple[torch.Tensor],
-        Optional[Tuple[torch.Tensor, Tuple[torch.FloatTensor, ...]]],
-    ]:
-        residual = hidden_states
-        hidden_states = self.ln_1(hidden_states)
-        attn_outputs = self.attn(
-            hidden_states,
-            layer_past=layer_past,
-            attention_mask=attention_mask,
-            head_mask=head_mask,
-            use_cache=use_cache,
-            output_attentions=output_attentions,
-        )
-        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
-        outputs = attn_outputs[1:]
-        # residual connection
-        hidden_states = attn_output + residual
-
-        if encoder_hidden_states is not None:
-            # add one self-attention block for cross-attention
-            if not hasattr(self, "crossattention"):
-                raise ValueError(
-                    f"If `encoder_hidden_states` are passed, {self} has to be instantiated with "
-                    "cross-attention layers by setting `config.add_cross_attention=True`"
-                )
-            residual = hidden_states
-            hidden_states = self.ln_cross_attn(hidden_states)
-            cross_attn_outputs = self.crossattention(
-                hidden_states,
-                attention_mask=attention_mask,
-                head_mask=head_mask,
-                encoder_hidden_states=encoder_hidden_states,
-                encoder_attention_mask=encoder_attention_mask,
-                output_attentions=output_attentions,
-            )
-            attn_output = cross_attn_outputs[0]
-            # residual connection
-            hidden_states = residual + attn_output
-            outputs = (
-                outputs + cross_attn_outputs[2:]
-            )  # add cross attentions if we output attention weights
-
-        residual = hidden_states
-        hidden_states = self.ln_2(hidden_states)
-        feed_forward_hidden_states = self.mlp(hidden_states)
-        # residual connection
-        hidden_states = residual + feed_forward_hidden_states
-
-        if use_cache:
-            outputs = (hidden_states,) + outputs
-        else:
-            outputs = (hidden_states,) + outputs[1:]
-
-        return outputs  # hidden_states, present, (attentions, cross_attentions)
-
-
-class GPT2PreTrainedModel(PreTrainedModel):
-    """
-    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
-    models.
-    """
-
-    config_class = GPT2Config
-    load_tf_weights = load_tf_weights_in_gpt2
-    base_model_prefix = "transformer"
-    is_parallelizable = True
-    supports_gradient_checkpointing = True
-    _no_split_modules = ["GPT2Block"]
-    _skip_keys_device_placement = "past_key_values"
-    _supports_flash_attn_2 = True
-    _supports_sdpa = True
-
-    def __init__(self, *inputs, **kwargs):
-        super().__init__(*inputs, **kwargs)
-
-    def _init_weights(self, module):
-        """Initialize the weights."""
-        if isinstance(module, (nn.Linear, Conv1D)):
-            # Slightly different from the TF version which uses truncated_normal for initialization
-            # cf https://github.com/pytorch/pytorch/pull/5617
-            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
-            if module.bias is not None:
-                module.bias.data.zero_()
-        elif isinstance(module, nn.Embedding):
-            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
-            if module.padding_idx is not None:
-                module.weight.data[module.padding_idx].zero_()
-        elif isinstance(module, nn.LayerNorm):
-            module.bias.data.zero_()
-            module.weight.data.fill_(1.0)
-
-        # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
-        #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
-        #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
-        #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
-        #
-        # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
-        for name, p in module.named_parameters():
-            if name == "c_proj.weight":
-                # Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
-                p.data.normal_(
-                    mean=0.0,
-                    std=(
-                        self.config.initializer_range
-                        / math.sqrt(2 * self.config.n_layer)
-                    ),
-                )
-
-
-@dataclass
-class GPT2DoubleHeadsModelOutput(ModelOutput):
-    """
-    Base class for outputs of models predicting if two sentences are consecutive or not.
-
-    Args:
-        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
-            Language modeling loss.
-        mc_loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `mc_labels` is provided):
-            Multiple choice classification loss.
-        logits (`torch.FloatTensor` of shape `(batch_size, num_choices, sequence_length, config.vocab_size)`):
-            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
-        mc_logits (`torch.FloatTensor` of shape `(batch_size, num_choices)`):
-            Prediction scores of the multiple choice classification head (scores for each choice before SoftMax).
-        past_key_values (`Tuple[Tuple[torch.Tensor]]`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
-            Tuple of length `config.n_layers`, containing tuples of tensors of shape `(batch_size, num_heads,
-            sequence_length, embed_size_per_head)`).
-
-            Contains pre-computed hidden-states (key and values in the attention blocks) that can be used (see
-            `past_key_values` input) to speed up sequential decoding.
-        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
-            Tuple of `torch.FloatTensor` (one for the output of the embeddings + one for the output of each layer) of
-            shape `(batch_size, sequence_length, hidden_size)`.
-
-            Hidden-states of the model at the output of each layer plus the initial embedding outputs.
-        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
-            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
-            sequence_length)`.
-
-            GPT2Attentions weights after the attention softmax, used to compute the weighted average in the
-            self-attention heads.
-    """
-
-    loss: Optional[torch.FloatTensor] = None
-    mc_loss: Optional[torch.FloatTensor] = None
-    logits: torch.FloatTensor = None
-    mc_logits: torch.FloatTensor = None
-    past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None
-    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
-    attentions: Optional[Tuple[torch.FloatTensor]] = None
-
-
-GPT2_START_DOCSTRING = r"""
-
-    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
-    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
-    etc.)
-
-    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
-    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
-    and behavior.
-
-    Parameters:
-        config ([`GPT2Config`]): 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.
-"""
-
-GPT2_INPUTS_DOCSTRING = r"""
-    Args:
-        input_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`):
-            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else
-            `past_key_values[0][0].shape[-2]` (`sequence_length` of input past key value states). Indices of input
-            sequence tokens in the vocabulary.
-
-            If `past_key_values` is used, only `input_ids` that do not have their past calculated should be passed as
-            `input_ids`.
-
-            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
-            [`PreTrainedTokenizer.__call__`] for details.
-
-            [What are input IDs?](../glossary#input-ids)
-        past_key_values (`Tuple[Tuple[torch.Tensor]]` of length `config.n_layers`):
-            Contains precomputed hidden-states (key and values in the attention blocks) as computed by the model (see
-            `past_key_values` output below). Can be used to speed up sequential decoding. The `input_ids` which have
-            their past given to this model should not be passed as `input_ids` as they have already been computed.
-        attention_mask (`torch.FloatTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
-
-            - 1 for tokens that are **not masked**,
-            - 0 for tokens that are **masked**.
-
-            If `past_key_values` is used, `attention_mask` needs to contain the masking strategy that was used for
-            `past_key_values`. In other words, the `attention_mask` always has to have the length:
-            `len(past_key_values) + len(input_ids)`
-
-            [What are attention masks?](../glossary#attention-mask)
-        token_type_ids (`torch.LongTensor` of shape `(batch_size, input_ids_length)`, *optional*):
-            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0,
-            1]`:
-
-            - 0 corresponds to a *sentence A* token,
-            - 1 corresponds to a *sentence B* token.
-
-            [What are token type IDs?](../glossary#token-type-ids)
-        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
-            config.max_position_embeddings - 1]`.
-
-            [What are position IDs?](../glossary#position-ids)
-        head_mask (`torch.FloatTensor` of shape `(num_heads,)` or `(num_layers, num_heads)`, *optional*):
-            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
-
-            - 1 indicates the head is **not masked**,
-            - 0 indicates the head is **masked**.
-
-        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
-            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
-            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
-            model's internal embedding lookup matrix.
-
-            If `past_key_values` is used, optionally only the last `inputs_embeds` have to be input (see
-            `past_key_values`).
-        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`).
-        output_attentions (`bool`, *optional*):
-            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
-            tensors for more detail.
-        output_hidden_states (`bool`, *optional*):
-            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
-            more detail.
-        return_dict (`bool`, *optional*):
-            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
-"""
-PARALLELIZE_DOCSTRING = r"""
-    This is an experimental feature and is a subject to change at a moment's notice.
-
-    Uses a device map to distribute attention modules of the model across several devices. If no device map is given,
-    it will evenly distribute blocks across all devices.
-
-    Args:
-        device_map (`Dict[int, list]`, *optional*):
-            A dictionary that maps attention modules to devices. Note that the embedding module and LMHead are always
-            automatically mapped to the first device (for esoteric reasons). That means that the first device should
-            have fewer attention modules mapped to it than other devices. For reference, the gpt2 models have the
-            following number of attention modules:
-
-                - openai-community/gpt2: 12
-                - openai-community/gpt2-medium: 24
-                - openai-community/gpt2-large: 36
-                - openai-community/gpt2-xl: 48
-
-    Example:
-
-    ```python
-    # Here is an example of a device map on a machine with 4 GPUs using gpt2-xl, which has a total of 48 attention modules:
-    model = GPT2LMHeadModel.from_pretrained("openai-community/gpt2-xl")
-    device_map = {
-        0: [0, 1, 2, 3, 4, 5, 6, 7, 8],
-        1: [9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21],
-        2: [22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34],
-        3: [35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47],
-    }
-    model.parallelize(device_map)
-    ```
-"""
-DEPARALLELIZE_DOCSTRING = r"""
-    Moves the model to cpu from a model parallel state.
-
-    Example:
-
-    ```python
-    # On a 4 GPU machine with openai-community/gpt2-large:
-    model = GPT2LMHeadModel.from_pretrained("openai-community/gpt2-large")
-    device_map = {
-        0: [0, 1, 2, 3, 4, 5, 6, 7],
-        1: [8, 9, 10, 11, 12, 13, 14, 15],
-        2: [16, 17, 18, 19, 20, 21, 22, 23],
-        3: [24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35],
-    }
-    model.parallelize(device_map)  # Splits the model across several devices
-    model.deparallelize()  # Put the model back on cpu and cleans memory by calling torch.cuda.empty_cache()
-    ```
-"""
-
-
-@add_start_docstrings(
-    "The bare GPT2 Model transformer outputting raw hidden-states without any specific head on top.",
-    GPT2_START_DOCSTRING,
-)
-class GPT2Model(GPT2PreTrainedModel):
-    _supports_param_buffer_assignment = False
-
-    def __init__(self, config):
-        super().__init__(config)
-
-        self.embed_dim = config.hidden_size
-
-        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
-        self.wpe = nn.Embedding(config.max_position_embeddings, self.embed_dim)
-
-        self.drop = nn.Dropout(config.embd_pdrop)
-        self.h = nn.ModuleList(
-            [GPT2Block(config, layer_idx=i) for i in range(config.num_hidden_layers)]
-        )
-        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
-
-        # Model parallel
-        self.model_parallel = False
-        self.device_map = None
-        self.gradient_checkpointing = False
-        self._attn_implementation = config._attn_implementation
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    @add_start_docstrings(PARALLELIZE_DOCSTRING)
-    def parallelize(self, device_map=None):
-        # Check validity of device_map
-        warnings.warn(
-            "`GPT2Model.parallelize` is deprecated and will be removed in v5 of Transformers, you should load your"
-            " model with `device_map='balanced'` in the call to `from_pretrained`. You can also provide your own"
-            " `device_map` but it needs to be a dictionary module_name to device, so for instance {'h.0': 0, 'h.1': 1,"
-            " ...}",
-            FutureWarning,
-        )
-        self.device_map = (
-            get_device_map(len(self.h), range(torch.cuda.device_count()))
-            if device_map is None
-            else device_map
-        )
-        assert_device_map(self.device_map, len(self.h))
-        self.model_parallel = True
-        self.first_device = (
-            "cpu"
-            if "cpu" in self.device_map.keys()
-            else "cuda:" + str(min(self.device_map.keys()))
-        )
-        self.last_device = "cuda:" + str(max(self.device_map.keys()))
-        self.wte = self.wte.to(self.first_device)
-        self.wpe = self.wpe.to(self.first_device)
-        # Load onto devices
-        for k, v in self.device_map.items():
-            for block in v:
-                cuda_device = "cuda:" + str(k)
-                self.h[block] = self.h[block].to(cuda_device)
-        # ln_f to last
-        self.ln_f = self.ln_f.to(self.last_device)
-
-    @add_start_docstrings(DEPARALLELIZE_DOCSTRING)
-    def deparallelize(self):
-        warnings.warn(
-            "Like `parallelize`, `deparallelize` is deprecated and will be removed in v5 of Transformers.",
-            FutureWarning,
-        )
-        self.model_parallel = False
-        self.device_map = None
-        self.first_device = "cpu"
-        self.last_device = "cpu"
-        self.wte = self.wte.to("cpu")
-        self.wpe = self.wpe.to("cpu")
-        for index in range(len(self.h)):
-            self.h[index] = self.h[index].to("cpu")
-        self.ln_f = self.ln_f.to("cpu")
-        torch.cuda.empty_cache()
-
-    def get_input_embeddings(self):
-        return self.wte
-
-    def set_input_embeddings(self, new_embeddings):
-        self.wte = new_embeddings
-
-    def _prune_heads(self, heads_to_prune):
-        """
-        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
-        """
-        for layer, heads in heads_to_prune.items():
-            self.h[layer].attn.prune_heads(heads)
-
-    @add_start_docstrings_to_model_forward(GPT2_INPUTS_DOCSTRING)
-    @add_code_sample_docstrings(
-        checkpoint=_CHECKPOINT_FOR_DOC,
-        output_type=BaseModelOutputWithPastAndCrossAttentions,
-        config_class=_CONFIG_FOR_DOC,
-    )
-    def forward(
-        self,
-        input_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        head_mask: Optional[torch.FloatTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        encoder_hidden_states: Optional[torch.Tensor] = None,
-        encoder_attention_mask: Optional[torch.FloatTensor] = None,
-        use_cache: Optional[bool] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, BaseModelOutputWithPastAndCrossAttentions]:
-        output_attentions = (
-            output_attentions
-            if output_attentions is not None
-            else self.config.output_attentions
-        )
-        output_hidden_states = (
-            output_hidden_states
-            if output_hidden_states is not None
-            else self.config.output_hidden_states
-        )
-        use_cache = use_cache if use_cache is not None else self.config.use_cache
-        return_dict = (
-            return_dict if return_dict is not None else self.config.use_return_dict
-        )
-
-        if input_ids is not None and inputs_embeds is not None:
-            raise ValueError(
-                "You cannot specify both input_ids and inputs_embeds at the same time"
-            )
-        elif input_ids is not None:
-            self.warn_if_padding_and_no_attention_mask(input_ids, attention_mask)
-            input_shape = input_ids.size()
-            input_ids = input_ids.view(-1, input_shape[-1])
-            batch_size = input_ids.shape[0]
-        elif inputs_embeds is not None:
-            input_shape = inputs_embeds.size()[:-1]
-            batch_size = inputs_embeds.shape[0]
-        else:
-            raise ValueError("You have to specify either input_ids or inputs_embeds")
-
-        device = input_ids.device if input_ids is not None else inputs_embeds.device
-
-        if token_type_ids is not None:
-            token_type_ids = token_type_ids.view(-1, input_shape[-1])
-
-        if past_key_values is None:
-            past_length = 0
-            past_key_values = tuple([None] * len(self.h))
-        else:
-            past_length = past_key_values[0][0].size(-2)
-        if position_ids is None:
-            position_ids = torch.arange(
-                past_length,
-                input_shape[-1] + past_length,
-                dtype=torch.long,
-                device=device,
-            )
-            position_ids = position_ids.unsqueeze(0)
-
-        if inputs_embeds is None:
-            inputs_embeds = self.wte(input_ids)
-        position_embeds = self.wpe(position_ids)
-        hidden_states = inputs_embeds + position_embeds
-
-        # Attention mask.
-        _use_sdpa = (
-            self._attn_implementation == "sdpa"
-            and output_attentions is False
-            and head_mask is None
-        )
-        attention_mask = (
-            attention_mask.view(batch_size, -1) if attention_mask is not None else None
-        )
-        if self._attn_implementation == "flash_attention_2":
-            attention_mask = (
-                attention_mask
-                if (attention_mask is not None and 0 in attention_mask)
-                else None
-            )
-        elif _use_sdpa:
-            attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
-                attention_mask=attention_mask,
-                input_shape=(batch_size, input_shape[-1]),
-                inputs_embeds=inputs_embeds,
-                past_key_values_length=past_length,
-            )
-        else:
-            if attention_mask is not None:
-                # We create a 3D attention mask from a 2D tensor mask.
-                # Sizes are [batch_size, 1, 1, to_seq_length]
-                # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
-                # this attention mask is more simple than the triangular masking of causal attention
-                # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
-                attention_mask = attention_mask[:, None, None, :]
-
-                # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
-                # masked positions, this operation will create a tensor which is 0.0 for
-                # positions we want to attend and the dtype's smallest value for masked positions.
-                # Since we are adding it to the raw scores before the softmax, this is
-                # effectively the same as removing these entirely.
-                attention_mask = attention_mask.to(
-                    dtype=self.dtype
-                )  # fp16 compatibility
-                attention_mask = (1.0 - attention_mask) * torch.finfo(self.dtype).min
-
-        # If a 2D or 3D attention mask is provided for the cross-attention
-        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
-        if self.config.add_cross_attention and encoder_hidden_states is not None:
-            encoder_batch_size, encoder_sequence_length, _ = (
-                encoder_hidden_states.size()
-            )
-            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
-            if encoder_attention_mask is None:
-                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
-            if _use_sdpa:
-                encoder_attention_mask = _prepare_4d_attention_mask_for_sdpa(
-                    mask=encoder_attention_mask,
-                    dtype=inputs_embeds.dtype,
-                    tgt_len=input_shape[-1],
-                )
-            elif not self._attn_implementation == "flash_attention_2":
-                encoder_attention_mask = self.invert_attention_mask(
-                    encoder_attention_mask
-                )
-        else:
-            encoder_attention_mask = None
-
-        # Prepare head mask if needed
-        # 1.0 in head_mask indicate we keep the head
-        # attention_probs has shape bsz x n_heads x N x N
-        # head_mask has shape n_layer x batch x n_heads x N x N
-        head_mask = self.get_head_mask(head_mask, self.config.n_layer)
-
-        if token_type_ids is not None:
-            token_type_embeds = self.wte(token_type_ids)
-            hidden_states = hidden_states + token_type_embeds
-
-        hidden_states = self.drop(hidden_states)
-
-        output_shape = (-1,) + input_shape[1:] + (hidden_states.size(-1),)
-
-        if self.gradient_checkpointing and self.training:
-            if use_cache:
-                logger.warning_once(
-                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
-                )
-                use_cache = False
-
-        presents = () if use_cache else None
-        all_self_attentions = () if output_attentions else None
-        all_cross_attentions = (
-            () if output_attentions and self.config.add_cross_attention else None
-        )
-        all_hidden_states = () if output_hidden_states else None
-        for i in range(len(self.h)):
-            block, layer_past = self.h[i], past_key_values[i]
-            # Model parallel
-            if self.model_parallel:
-                torch.cuda.set_device(hidden_states.device)
-                # Ensure layer_past is on same device as hidden_states (might not be correct)
-                if layer_past is not None:
-                    layer_past = tuple(
-                        past_state.to(hidden_states.device) for past_state in layer_past
-                    )
-                # Ensure that attention_mask is always on the same device as hidden_states
-                if attention_mask is not None:
-                    attention_mask = attention_mask.to(hidden_states.device)
-                if isinstance(head_mask, torch.Tensor):
-                    head_mask = head_mask.to(hidden_states.device)
-            if output_hidden_states:
-                all_hidden_states = all_hidden_states + (hidden_states,)
-
-            if self.gradient_checkpointing and self.training:
-                outputs = self._gradient_checkpointing_func(
-                    block.__call__,
-                    hidden_states,
-                    None,
-                    attention_mask,
-                    head_mask[i],
-                    encoder_hidden_states,
-                    encoder_attention_mask,
-                    use_cache,
-                    output_attentions,
-                )
-            else:
-                outputs = block(
-                    hidden_states,
-                    layer_past=layer_past,
-                    attention_mask=attention_mask,
-                    head_mask=head_mask[i],
-                    encoder_hidden_states=encoder_hidden_states,
-                    encoder_attention_mask=encoder_attention_mask,
-                    use_cache=use_cache,
-                    output_attentions=output_attentions,
-                )
-
-            hidden_states = outputs[0]
-            if use_cache is True:
-                presents = presents + (outputs[1],)
-
-            if output_attentions:
-                all_self_attentions = all_self_attentions + (
-                    outputs[2 if use_cache else 1],
-                )
-                if self.config.add_cross_attention:
-                    all_cross_attentions = all_cross_attentions + (
-                        outputs[3 if use_cache else 2],
-                    )
-
-            # Model Parallel: If it's the last layer for that device, put things on the next device
-            if self.model_parallel:
-                for k, v in self.device_map.items():
-                    if i == v[-1] and "cuda:" + str(k) != self.last_device:
-                        hidden_states = hidden_states.to("cuda:" + str(k + 1))
-
-        hidden_states = self.ln_f(hidden_states)
-
-        hidden_states = hidden_states.view(output_shape)
-        # Add last hidden state
-        if output_hidden_states:
-            all_hidden_states = all_hidden_states + (hidden_states,)
-
-        if not return_dict:
-            return tuple(
-                v
-                for v in [
-                    hidden_states,
-                    presents,
-                    all_hidden_states,
-                    all_self_attentions,
-                    all_cross_attentions,
-                ]
-                if v is not None
-            )
-
-        return BaseModelOutputWithPastAndCrossAttentions(
-            last_hidden_state=hidden_states,
-            past_key_values=presents,
-            hidden_states=all_hidden_states,
-            attentions=all_self_attentions,
-            cross_attentions=all_cross_attentions,
-        )
-
-
-@add_start_docstrings(
-    """
-    The GPT2 Model transformer with a language modeling head on top (linear layer with weights tied to the input
-    embeddings).
-    """,
-    GPT2_START_DOCSTRING,
-)
-class GPT2LMHeadModel(GPT2PreTrainedModel, GenerationMixin):
-    _tied_weights_keys = ["lm_head.weight"]
-
-    def __init__(self, config):
-        super().__init__(config)
-        self.transformer = GPT2Model(config)
-        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
-
-        # Model parallel
-        self.model_parallel = False
-        self.device_map = None
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    @add_start_docstrings(PARALLELIZE_DOCSTRING)
-    def parallelize(self, device_map=None):
-        warnings.warn(
-            "`GPT2LMHeadModel.parallelize` is deprecated and will be removed in v5 of Transformers, you should load"
-            " your model with `device_map='balanced'` in the call to `from_pretrained`. You can also provide your own"
-            " `device_map` but it needs to be a dictionary module_name to device, so for instance {'transformer.h.0':"
-            " 0, 'transformer.h.1': 1, ...}",
-            FutureWarning,
-        )
-        self.device_map = (
-            get_device_map(len(self.transformer.h), range(torch.cuda.device_count()))
-            if device_map is None
-            else device_map
-        )
-        assert_device_map(self.device_map, len(self.transformer.h))
-        self.transformer.parallelize(self.device_map)
-        self.lm_head = self.lm_head.to(self.transformer.first_device)
-        self.model_parallel = True
-
-    @add_start_docstrings(DEPARALLELIZE_DOCSTRING)
-    def deparallelize(self):
-        warnings.warn(
-            "Like `parallelize`, `deparallelize` is deprecated and will be removed in v5 of Transformers.",
-            FutureWarning,
-        )
-        self.transformer.deparallelize()
-        self.transformer = self.transformer.to("cpu")
-        self.lm_head = self.lm_head.to("cpu")
-        self.model_parallel = False
-        torch.cuda.empty_cache()
-
-    def get_output_embeddings(self):
-        return self.lm_head
-
-    def set_output_embeddings(self, new_embeddings):
-        self.lm_head = new_embeddings
-
-    @add_start_docstrings_to_model_forward(GPT2_INPUTS_DOCSTRING)
-    @add_code_sample_docstrings(
-        checkpoint=_CHECKPOINT_FOR_DOC,
-        output_type=CausalLMOutputWithCrossAttentions,
-        config_class=_CONFIG_FOR_DOC,
-    )
-    def forward(
-        self,
-        input_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        head_mask: Optional[torch.FloatTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        encoder_hidden_states: Optional[torch.Tensor] = None,
-        encoder_attention_mask: 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,
-        return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, CausalLMOutputWithCrossAttentions]:
-        r"""
-        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
-            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
-            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
-        """
-        return_dict = (
-            return_dict if return_dict is not None else self.config.use_return_dict
-        )
-
-        transformer_outputs = self.transformer(
-            input_ids,
-            past_key_values=past_key_values,
-            attention_mask=attention_mask,
-            token_type_ids=token_type_ids,
-            position_ids=position_ids,
-            head_mask=head_mask,
-            inputs_embeds=inputs_embeds,
-            encoder_hidden_states=encoder_hidden_states,
-            encoder_attention_mask=encoder_attention_mask,
-            use_cache=use_cache,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-        )
-        hidden_states = transformer_outputs[0]
-
-        # Set device for model parallelism
-        if self.model_parallel:
-            torch.cuda.set_device(self.transformer.first_device)
-            hidden_states = hidden_states.to(self.lm_head.weight.device)
-
-        lm_logits = self.lm_head(hidden_states)
-
-        loss = None
-        if labels is not None:
-            # move labels to correct device to enable model parallelism
-            labels = labels.to(lm_logits.device)
-            # Shift so that tokens < n predict n
-            shift_logits = lm_logits[..., :-1, :].contiguous()
-            shift_labels = labels[..., 1:].contiguous()
-            # Flatten the tokens
-            loss_fct = CrossEntropyLoss()
-            loss = loss_fct(
-                shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1)
-            )
-
-        if not return_dict:
-            output = (lm_logits,) + transformer_outputs[1:]
-            return ((loss,) + output) if loss is not None else output
-
-        return CausalLMOutputWithCrossAttentions(
-            loss=loss,
-            logits=lm_logits,
-            past_key_values=transformer_outputs.past_key_values,
-            hidden_states=transformer_outputs.hidden_states,
-            attentions=transformer_outputs.attentions,
-            cross_attentions=transformer_outputs.cross_attentions,
-        )
-
-    @staticmethod
-    def _reorder_cache(
-        past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
-    ) -> Tuple[Tuple[torch.Tensor]]:
-        """
-        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
-        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
-        beam_idx at every generation step.
-        """
-        return tuple(
-            tuple(
-                past_state.index_select(0, beam_idx.to(past_state.device))
-                for past_state in layer_past
-            )
-            for layer_past in past_key_values
-        )
-
-
-@add_start_docstrings(
-    """
-The GPT2 Model transformer with a language modeling and a multiple-choice classification head on top e.g. for
-RocStories/SWAG tasks. The two heads are two linear layers. The language modeling head has its weights tied to the
-input embeddings, the classification head takes as input the input of a specified classification token index in the
-input sequence).
-""",
-    GPT2_START_DOCSTRING,
-)
-class GPT2DoubleHeadsModel(GPT2PreTrainedModel, GenerationMixin):
-    _tied_weights_keys = ["lm_head.weight"]
-
-    def __init__(self, config):
-        super().__init__(config)
-        config.num_labels = 1
-        self.transformer = GPT2Model(config)
-        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
-        self.multiple_choice_head = SequenceSummary(config)
-
-        # Model parallel
-        self.model_parallel = False
-        self.device_map = None
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    @add_start_docstrings(PARALLELIZE_DOCSTRING)
-    def parallelize(self, device_map=None):
-        warnings.warn(
-            "`GPT2DoubleHeadsModel.parallelize` is deprecated and will be removed in v5 of Transformers, you should"
-            " load your model with `device_map='balanced'` in the call to `from_pretrained`. You can also provide your"
-            " own `device_map` but it needs to be a dictionary module_name to device, so for instance"
-            " {'transformer.h.0': 0, 'transformer.h.1': 1, ...}",
-            FutureWarning,
-        )
-        self.device_map = (
-            get_device_map(len(self.transformer.h), range(torch.cuda.device_count()))
-            if device_map is None
-            else device_map
-        )
-        assert_device_map(self.device_map, len(self.transformer.h))
-        self.transformer.parallelize(self.device_map)
-        self.lm_head = self.lm_head.to(self.transformer.first_device)
-        self.multiple_choice_head = self.multiple_choice_head.to(
-            self.transformer.first_device
-        )
-        self.model_parallel = True
-
-    @add_start_docstrings(DEPARALLELIZE_DOCSTRING)
-    def deparallelize(self):
-        warnings.warn(
-            "Like `parallelize`, `deparallelize` is deprecated and will be removed in v5 of Transformers.",
-            FutureWarning,
-        )
-        self.transformer.deparallelize()
-        self.transformer = self.transformer.to("cpu")
-        self.lm_head = self.lm_head.to("cpu")
-        self.multiple_choice_head = self.multiple_choice_head.to("cpu")
-        self.model_parallel = False
-        torch.cuda.empty_cache()
-
-    def get_output_embeddings(self):
-        return self.lm_head
-
-    def set_output_embeddings(self, new_embeddings):
-        self.lm_head = new_embeddings
-
-    @add_start_docstrings_to_model_forward(GPT2_INPUTS_DOCSTRING)
-    @replace_return_docstrings(
-        output_type=GPT2DoubleHeadsModelOutput, config_class=_CONFIG_FOR_DOC
-    )
-    def forward(
-        self,
-        input_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        head_mask: Optional[torch.FloatTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        mc_token_ids: Optional[torch.LongTensor] = None,
-        labels: Optional[torch.LongTensor] = None,
-        mc_labels: Optional[torch.LongTensor] = None,
-        use_cache: Optional[bool] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-        **kwargs,
-    ) -> Union[Tuple, GPT2DoubleHeadsModelOutput]:
-        r"""
-        mc_token_ids (`torch.LongTensor` of shape `(batch_size, num_choices)`, *optional*, default to index of the last token of the input):
-            Index of the classification token in each input sequence. Selected in the range `[0, input_ids.size(-1) -
-            1]`.
-        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
-            `labels = input_ids`. Indices are selected in `[-100, 0, ..., config.vocab_size - 1]`. All labels set to
-            `-100` are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size - 1]`
-        mc_labels (`torch.LongTensor` of shape `(batch_size)`, *optional*):
-            Labels for computing the multiple choice classification loss. Indices should be in `[0, ..., num_choices]`
-            where *num_choices* is the size of the second dimension of the input tensors. (see *input_ids* above)
-
-        Return:
-
-        Example:
-
-        ```python
-        >>> import torch
-        >>> from transformers import AutoTokenizer, GPT2DoubleHeadsModel
-
-        >>> tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2")
-        >>> model = GPT2DoubleHeadsModel.from_pretrained("openai-community/gpt2")
-
-        >>> # Add a [CLS] to the vocabulary (we should train it also!)
-        >>> num_added_tokens = tokenizer.add_special_tokens({"cls_token": "[CLS]"})
-        >>> # Update the model embeddings with the new vocabulary size
-        >>> embedding_layer = model.resize_token_embeddings(len(tokenizer))
-
-        >>> choices = ["Hello, my dog is cute [CLS]", "Hello, my cat is cute [CLS]"]
-        >>> encoded_choices = [tokenizer.encode(s) for s in choices]
-        >>> cls_token_location = [tokens.index(tokenizer.cls_token_id) for tokens in encoded_choices]
-
-        >>> input_ids = torch.tensor(encoded_choices).unsqueeze(0)  # Batch size: 1, number of choices: 2
-        >>> mc_token_ids = torch.tensor([cls_token_location])  # Batch size: 1
-
-        >>> outputs = model(input_ids, mc_token_ids=mc_token_ids)
-        >>> lm_logits = outputs.logits
-        >>> mc_logits = outputs.mc_logits
-        ```"""
-        return_dict = (
-            return_dict if return_dict is not None else self.config.use_return_dict
-        )
-
-        transformer_outputs = self.transformer(
-            input_ids,
-            past_key_values=past_key_values,
-            attention_mask=attention_mask,
-            token_type_ids=token_type_ids,
-            position_ids=position_ids,
-            head_mask=head_mask,
-            inputs_embeds=inputs_embeds,
-            use_cache=use_cache,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-        )
-
-        hidden_states = transformer_outputs[0]
-
-        # Set device for model parallelism
-        if self.model_parallel:
-            torch.cuda.set_device(self.transformer.first_device)
-            hidden_states = hidden_states.to(self.lm_head.weight.device)
-
-        lm_logits = self.lm_head(hidden_states)
-        mc_logits = self.multiple_choice_head(hidden_states, mc_token_ids).squeeze(-1)
-
-        mc_loss = None
-        if mc_labels is not None:
-            loss_fct = CrossEntropyLoss()
-            mc_loss = loss_fct(
-                mc_logits.view(-1, mc_logits.size(-1)), mc_labels.view(-1)
-            )
-        lm_loss = None
-        if labels is not None:
-            labels = labels.to(lm_logits.device)
-            shift_logits = lm_logits[..., :-1, :].contiguous()
-            shift_labels = labels[..., 1:].contiguous()
-            loss_fct = CrossEntropyLoss()
-            lm_loss = loss_fct(
-                shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1)
-            )
-
-        if not return_dict:
-            output = (lm_logits, mc_logits) + transformer_outputs[1:]
-            if mc_loss is not None:
-                output = (mc_loss,) + output
-            return ((lm_loss,) + output) if lm_loss is not None else output
-
-        return GPT2DoubleHeadsModelOutput(
-            loss=lm_loss,
-            mc_loss=mc_loss,
-            logits=lm_logits,
-            mc_logits=mc_logits,
-            past_key_values=transformer_outputs.past_key_values,
-            hidden_states=transformer_outputs.hidden_states,
-            attentions=transformer_outputs.attentions,
-        )
-
-    @staticmethod
-    def _reorder_cache(
-        past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
-    ) -> Tuple[Tuple[torch.Tensor]]:
-        """
-        This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
-        [`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
-        beam_idx at every generation step.
-        """
-        return tuple(
-            tuple(
-                past_state.index_select(0, beam_idx.to(past_state.device))
-                for past_state in layer_past
-            )
-            for layer_past in past_key_values
-        )
-
-
-@add_start_docstrings(
-    """
-    The GPT2 Model transformer with a sequence classification head on top (linear layer).
-
-    [`GPT2ForSequenceClassification`] uses the last token in order to do the classification, as other causal models
-    (e.g. GPT-1) do.
-
-    Since it does classification on the last token, it requires to know the position of the last token. If a
-    `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
-    no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
-    padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
-    each row of the batch).
-    """,
-    GPT2_START_DOCSTRING,
-)
-class GPT2ForSequenceClassification(GPT2PreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-        self.num_labels = config.num_labels
-        self.transformer = GPT2Model(config)
-        self.score = nn.Linear(config.n_embd, self.num_labels, bias=False)
-
-        # Model parallel
-        self.model_parallel = False
-        self.device_map = None
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    @add_start_docstrings_to_model_forward(GPT2_INPUTS_DOCSTRING)
-    @add_code_sample_docstrings(
-        checkpoint="microsoft/DialogRPT-updown",
-        output_type=SequenceClassifierOutputWithPast,
-        config_class=_CONFIG_FOR_DOC,
-    )
-    def forward(
-        self,
-        input_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        head_mask: Optional[torch.FloatTensor] = 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,
-        return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
-        r"""
-        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
-            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
-            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
-            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
-        """
-        return_dict = (
-            return_dict if return_dict is not None else self.config.use_return_dict
-        )
-
-        transformer_outputs = self.transformer(
-            input_ids,
-            past_key_values=past_key_values,
-            attention_mask=attention_mask,
-            token_type_ids=token_type_ids,
-            position_ids=position_ids,
-            head_mask=head_mask,
-            inputs_embeds=inputs_embeds,
-            use_cache=use_cache,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-        )
-        hidden_states = transformer_outputs[0]
-        logits = self.score(hidden_states)
-
-        if input_ids is not None:
-            batch_size, sequence_length = input_ids.shape[:2]
-        else:
-            batch_size, sequence_length = inputs_embeds.shape[:2]
-
-        assert (
-            self.config.pad_token_id is not None or batch_size == 1
-        ), "Cannot handle batch sizes > 1 if no padding token is defined."
-        if self.config.pad_token_id is None:
-            sequence_lengths = -1
-        else:
-            if input_ids is not None:
-                # if no pad token found, use modulo instead of reverse indexing for ONNX compatibility
-                sequence_lengths = (
-                    torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1
-                )
-                sequence_lengths = sequence_lengths % input_ids.shape[-1]
-                sequence_lengths = sequence_lengths.to(logits.device)
-            else:
-                sequence_lengths = -1
-                logger.warning_once(
-                    f"{self.__class__.__name__} will not detect padding tokens in `inputs_embeds`. Results may be "
-                    "unexpected if using padding tokens in conjunction with `inputs_embeds.`"
-                )
-
-        pooled_logits = logits[
-            torch.arange(batch_size, device=logits.device), sequence_lengths
-        ]
-
-        loss = None
-        if labels is not None:
-            if self.config.problem_type is None:
-                if self.num_labels == 1:
-                    self.config.problem_type = "regression"
-                elif self.num_labels > 1 and (
-                    labels.dtype == torch.long or labels.dtype == torch.int
-                ):
-                    self.config.problem_type = "single_label_classification"
-                else:
-                    self.config.problem_type = "multi_label_classification"
-
-            if self.config.problem_type == "regression":
-                loss_fct = MSELoss()
-                if self.num_labels == 1:
-                    loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
-                else:
-                    loss = loss_fct(pooled_logits, labels)
-            elif self.config.problem_type == "single_label_classification":
-                loss_fct = CrossEntropyLoss()
-                loss = loss_fct(
-                    pooled_logits.view(-1, self.num_labels), labels.view(-1)
-                )
-            elif self.config.problem_type == "multi_label_classification":
-                loss_fct = BCEWithLogitsLoss()
-                loss = loss_fct(pooled_logits, labels)
-        if not return_dict:
-            output = (pooled_logits,) + transformer_outputs[1:]
-            return ((loss,) + output) if loss is not None else output
-
-        return SequenceClassifierOutputWithPast(
-            loss=loss,
-            logits=pooled_logits,
-            past_key_values=transformer_outputs.past_key_values,
-            hidden_states=transformer_outputs.hidden_states,
-            attentions=transformer_outputs.attentions,
-        )
-
-
-@add_start_docstrings(
-    """
-    GPT2 Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for
-    Named-Entity-Recognition (NER) tasks.
-    """,
-    GPT2_START_DOCSTRING,
-)
-class GPT2ForTokenClassification(GPT2PreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-        self.num_labels = config.num_labels
-
-        self.transformer = GPT2Model(config)
-        if (
-            hasattr(config, "classifier_dropout")
-            and config.classifier_dropout is not None
-        ):
-            classifier_dropout = config.classifier_dropout
-        elif hasattr(config, "hidden_dropout") and config.hidden_dropout is not None:
-            classifier_dropout = config.hidden_dropout
-        else:
-            classifier_dropout = 0.1
-        self.dropout = nn.Dropout(classifier_dropout)
-        self.classifier = nn.Linear(config.hidden_size, config.num_labels)
-
-        # Model parallel
-        self.model_parallel = False
-        self.device_map = None
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    @add_start_docstrings_to_model_forward(GPT2_INPUTS_DOCSTRING)
-    # fmt: off
-    @add_code_sample_docstrings(
-        checkpoint="brad1141/gpt2-finetuned-comp2",
-        output_type=TokenClassifierOutput,
-        config_class=_CONFIG_FOR_DOC,
-        expected_loss=0.25,
-        expected_output=[
-            "Lead",
-            "Lead",
-            "Lead",
-            "Position",
-            "Lead",
-            "Lead",
-            "Lead",
-            "Lead",
-            "Lead",
-            "Lead",
-            "Lead",
-            "Lead",
-        ],
-    )
-    # fmt: on
-    def forward(
-        self,
-        input_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
-        attention_mask: Optional[torch.FloatTensor] = None,
-        token_type_ids: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        head_mask: Optional[torch.FloatTensor] = 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,
-        return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, TokenClassifierOutput]:
-        r"""
-        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
-            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
-            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
-            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
-        """
-        return_dict = (
-            return_dict if return_dict is not None else self.config.use_return_dict
-        )
-
-        transformer_outputs = self.transformer(
-            input_ids,
-            past_key_values=past_key_values,
-            attention_mask=attention_mask,
-            token_type_ids=token_type_ids,
-            position_ids=position_ids,
-            head_mask=head_mask,
-            inputs_embeds=inputs_embeds,
-            use_cache=use_cache,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-        )
-
-        hidden_states = transformer_outputs[0]
-        hidden_states = self.dropout(hidden_states)
-        logits = self.classifier(hidden_states)
-
-        loss = None
-        if labels is not None:
-            labels = labels.to(logits.device)
-            loss_fct = CrossEntropyLoss()
-            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
-
-        if not return_dict:
-            output = (logits,) + transformer_outputs[2:]
-            return ((loss,) + output) if loss is not None else output
-
-        return TokenClassifierOutput(
-            loss=loss,
-            logits=logits,
-            hidden_states=transformer_outputs.hidden_states,
-            attentions=transformer_outputs.attentions,
-        )
-
-
-@add_start_docstrings(
-    """
-    The GPT-2 Model transformer with a span classification head on top for extractive question-answering tasks like
-    SQuAD (a linear layer on top of the hidden-states output to compute `span start logits` and `span end logits`).
-    """,
-    GPT2_START_DOCSTRING,
-)
-class GPT2ForQuestionAnswering(GPT2PreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-        self.num_labels = config.num_labels
-        self.transformer = GPT2Model(config)
-        self.qa_outputs = nn.Linear(config.hidden_size, 2)
-
-        # Model parallel
-        self.model_parallel = False
-        self.device_map = None
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    @add_start_docstrings_to_model_forward(
-        GPT2_INPUTS_DOCSTRING.format("batch_size, sequence_length")
-    )
-    @add_code_sample_docstrings(
-        checkpoint=_CHECKPOINT_FOR_DOC,
-        output_type=QuestionAnsweringModelOutput,
-        config_class=_CONFIG_FOR_DOC,
-        real_checkpoint=_CHECKPOINT_FOR_DOC,
-    )
-    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,
-        head_mask: Optional[torch.FloatTensor] = None,
-        inputs_embeds: Optional[torch.FloatTensor] = None,
-        start_positions: Optional[torch.LongTensor] = None,
-        end_positions: Optional[torch.LongTensor] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, QuestionAnsweringModelOutput]:
-        r"""
-        start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
-            Labels for position (index) of the start of the labelled span for computing the token classification loss.
-            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
-            are not taken into account for computing the loss.
-        end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
-            Labels for position (index) of the end of the labelled span for computing the token classification loss.
-            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
-            are not taken into account for computing the loss.
-        """
-        return_dict = (
-            return_dict if return_dict is not None else self.config.use_return_dict
-        )
-
-        outputs = self.transformer(
-            input_ids,
-            attention_mask=attention_mask,
-            token_type_ids=token_type_ids,
-            position_ids=position_ids,
-            head_mask=head_mask,
-            inputs_embeds=inputs_embeds,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-        )
-
-        sequence_output = outputs[0]
-
-        logits = self.qa_outputs(sequence_output)
-        start_logits, end_logits = logits.split(1, dim=-1)
-        start_logits = start_logits.squeeze(-1).contiguous()
-        end_logits = end_logits.squeeze(-1).contiguous()
-
-        total_loss = None
-        if start_positions is not None and end_positions is not None:
-            # If we are on multi-GPU, split add a dimension
-            if len(start_positions.size()) > 1:
-                start_positions = start_positions.squeeze(-1).to(start_logits.device)
-            if len(end_positions.size()) > 1:
-                end_positions = end_positions.squeeze(-1).to(end_logits.device)
-            # sometimes the start/end positions are outside our model inputs, we ignore these terms
-            ignored_index = start_logits.size(1)
-            start_positions = start_positions.clamp(0, ignored_index)
-            end_positions = end_positions.clamp(0, ignored_index)
-
-            loss_fct = CrossEntropyLoss(ignore_index=ignored_index)
-            start_loss = loss_fct(start_logits, start_positions)
-            end_loss = loss_fct(end_logits, end_positions)
-            total_loss = (start_loss + end_loss) / 2
-
-        if not return_dict:
-            output = (start_logits, end_logits) + outputs[2:]
-            return ((total_loss,) + output) if total_loss is not None else output
+        return attn_output, attn_weights
 
-        return QuestionAnsweringModelOutput(
-            loss=total_loss,
-            start_logits=start_logits,
-            end_logits=end_logits,
-            hidden_states=outputs.hidden_states,
-            attentions=outputs.attentions,
-        )
+__all__ = [
+    "GPT2DoubleHeadsModel",
+    "GPT2ForQuestionAnswering",
+    "GPT2ForSequenceClassification",
+    "GPT2ForTokenClassification",
+    "GPT2LMHeadModel",
+    "GPT2Model",
+    "GPT2PreTrainedModel",
+    "load_tf_weights_in_gpt2",
+]