#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/esmc/modular_esmc.py.
#               Do NOT edit this file manually as any edits will be overwritten by the generation of
#             the file from the modular. If any change should be done, please apply the change to the
#                          modular_esmc.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# Copyright 2026 BioHub and The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import math
from collections.abc import Callable

import torch
import torch.nn as nn
from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
from torch.nn import functional as F

from ...activations import ACT2FN
from ...integrations import use_kernel_forward_from_hub, use_kernelized_func
from ...masking_utils import create_bidirectional_mask, packed_sequence_mask_function
from ...modeling_layers import GenericForTokenClassification, GradientCheckpointingLayer
from ...modeling_outputs import BaseModelOutput, MaskedLMOutput, SequenceClassifierOutput
from ...modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
from ...modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
from ...processing_utils import Unpack
from ...utils import TransformersKwargs, auto_docstring, can_return_tuple
from ...utils.deprecation import deprecate_kwarg
from ...utils.generic import is_flash_attention_requested, maybe_autocast, merge_with_config_defaults
from ...utils.output_capturing import capture_outputs
from .configuration_esmc import EsmcConfig


class EsmcLayerNorm(nn.LayerNorm):
    """LayerNorm that returns its input dtype.

    The reference fuses each LayerNorm into the projection that follows it, keeping the fp32 reduction
    inside the fused op; a bare ``nn.LayerNorm`` under autocast instead hands back fp32 and promotes
    the residual stream, leaving the rotary ``cos``/``sin`` as the only bf16 tensors in the block.
    """

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        return super().forward(hidden_states).to(hidden_states.dtype)


class EsmcRotaryEmbedding(nn.Module):
    @deprecate_kwarg("device", version="5.18")
    def __init__(self, config: EsmcConfig, device=None):
        super().__init__()
        self.max_seq_len_cached = config.max_position_embeddings
        self.original_max_seq_len = config.max_position_embeddings

        self.config = config

        self.rope_type = self.config.rope_parameters["rope_type"]
        rope_init_fn: Callable = self.compute_default_rope_parameters
        if self.rope_type != "default":
            rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
        inv_freq, self.attention_scaling = rope_init_fn(self.config, device)

        self.inv_freq = nn.Buffer(inv_freq, persistent=False)
        self.original_inv_freq = nn.Buffer(inv_freq.clone(), persistent=False)

    @staticmethod
    @deprecate_kwarg("device", version="5.18")
    def compute_default_rope_parameters(config: EsmcConfig, device=None, **kwargs) -> tuple[torch.Tensor, float]:
        """
        Computes the inverse frequencies according to the original RoPE implementation
        Args:
            config ([`~transformers.PreTrainedConfig`]):
                The model configuration.
        Returns:
            Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
            post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE).
        """
        base = config.rope_parameters["rope_theta"]
        dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads

        attention_factor = 1.0  # Unused in this type of RoPE
        # Compute the inverse frequencies
        inv_freq = 1.0 / (base ** (torch.arange(0, dim, 2, dtype=torch.float) / dim))
        return inv_freq.to(device), attention_factor

    @torch.no_grad()
    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
    def forward(self, x, position_ids):
        inv_freq_expanded = (
            self.inv_freq[None, :, None].expand(position_ids.shape[0], -1, 1).to(dtype=torch.float, device=x.device)
        )
        position_ids_expanded = position_ids[:, None, :].float()

        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
        # Disable any outside autocast context if any, to really force fp32
        with maybe_autocast(device_type=device_type, enabled=False):
            freqs = (inv_freq_expanded @ position_ids_expanded).transpose(1, 2)
            emb = torch.cat((freqs, freqs), dim=-1)
            cos = emb.cos() * self.attention_scaling
            sin = emb.sin() * self.attention_scaling

        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)


class EsmcMLP(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.config = config
        self.hidden_size = config.hidden_size
        self.intermediate_size = config.intermediate_size
        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias)
        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=config.mlp_bias)
        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=config.mlp_bias)
        self.act_fn = ACT2FN[config.hidden_act]

    def forward(self, x):
        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
        return down_proj


def rotate_half(x):
    """Rotates half the hidden dims of the input."""
    x1 = x[..., : x.shape[-1] // 2]
    x2 = x[..., x.shape[-1] // 2 :]
    return torch.cat((-x2, x1), dim=-1)


@use_kernel_forward_from_hub("rotary_pos_emb")
def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1):
    """Applies Rotary Position Embedding to the query and key tensors.

    Args:
        q (`torch.Tensor`): The query tensor.
        k (`torch.Tensor`): The key tensor.
        cos (`torch.Tensor`): The cosine part of the rotary embedding.
        sin (`torch.Tensor`): The sine part of the rotary embedding.
        unsqueeze_dim (`int`, *optional*, defaults to 1):
            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
    Returns:
        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
    """
    cos = cos.unsqueeze(unsqueeze_dim)
    sin = sin.unsqueeze(unsqueeze_dim)
    q_embed = (q * cos) + (rotate_half(q) * sin)
    k_embed = (k * cos) + (rotate_half(k) * sin)
    return q_embed, k_embed


def eager_attention_forward(
    module: nn.Module,
    query: torch.Tensor,
    key: torch.Tensor,
    value: torch.Tensor,
    attention_mask: torch.Tensor | None,
    scaling: float | None = None,
    dropout: float = 0.0,
    **kwargs: Unpack[TransformersKwargs],
):
    if scaling is None:
        scaling = query.size(-1) ** -0.5

    # Take the dot product between "query" and "key" to get the raw attention scores.
    attn_weights = torch.matmul(query, key.transpose(2, 3)) * scaling

    if attention_mask is not None:
        attn_weights = attn_weights + attention_mask

    attn_weights = nn.functional.softmax(attn_weights, dim=-1)
    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)

    attn_output = torch.matmul(attn_weights, value)
    attn_output = attn_output.transpose(1, 2).contiguous()

    return attn_output, attn_weights


@use_kernelized_func(apply_rotary_pos_emb)
class EsmcAttention(nn.Module):
    """Multi-head self-attention with QK-LayerNorm and RoPE."""

    def __init__(self, config: EsmcConfig, layer_idx: int | None = None):
        super().__init__()
        self.config = config
        self.layer_idx = layer_idx
        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
        self.scaling = self.head_dim**-0.5
        self.attention_dropout = config.attention_dropout
        self.is_causal = False

        self.q_proj = nn.Linear(
            config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias
        )
        self.k_proj = nn.Linear(
            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
        )
        self.v_proj = nn.Linear(
            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
        )
        self.o_proj = nn.Linear(
            config.num_attention_heads * self.head_dim, config.hidden_size, bias=config.attention_bias
        )
        # Every released ESMC checkpoint carries q_norm/k_norm weights, so there is no flag for these.
        self.q_norm = EsmcLayerNorm(config.hidden_size, bias=False)
        self.k_norm = EsmcLayerNorm(config.hidden_size, bias=False)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: torch.Tensor | None = None,
        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple[torch.Tensor, torch.Tensor | None]:
        input_shape = hidden_states.shape[:-1]
        hidden_shape = (*input_shape, -1, self.head_dim)

        query_states = self.q_norm(self.q_proj(hidden_states)).to(hidden_states.dtype)
        key_states = self.k_norm(self.k_proj(hidden_states)).to(hidden_states.dtype)
        value_states = self.v_proj(hidden_states)

        query_states = query_states.view(hidden_shape).transpose(1, 2)
        key_states = key_states.view(hidden_shape).transpose(1, 2)
        value_states = value_states.view(hidden_shape).transpose(1, 2)

        cos, sin = position_embeddings
        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)

        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
            self.config._attn_implementation, eager_attention_forward
        )

        attn_output, attn_weights = attention_interface(
            self,
            query_states,
            key_states,
            value_states,
            attention_mask,
            dropout=0.0 if not self.training else self.attention_dropout,
            scaling=self.scaling,
            **kwargs,
        )
        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
        return self.o_proj(attn_output), attn_weights


class EsmcLayer(GradientCheckpointingLayer):
    """Single transformer block: pre-norm attention + pre-norm FFN with residual scaling."""

    def __init__(self, config: EsmcConfig, layer_idx: int | None = None):
        super().__init__()
        self.hidden_size = config.hidden_size

        self.self_attn = EsmcAttention(config=config, layer_idx=layer_idx)

        self.mlp = EsmcMLP(config)
        # LayerNorm instead of Llama's RMSNorm, dtype-restoring (see EsmcLayerNorm).
        self.input_layernorm = EsmcLayerNorm(config.hidden_size)
        self.post_attention_layernorm = EsmcLayerNorm(config.hidden_size)
        # ESM3 residual scaling to stabilise deep networks; every released checkpoint uses it.
        self.scaling_factor = math.sqrt(config.num_hidden_layers / 36)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: torch.Tensor | None = None,
        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> torch.Tensor:
        residual = hidden_states
        attn_output, _ = self.self_attn(
            self.input_layernorm(hidden_states),
            attention_mask,
            position_embeddings=position_embeddings,
            **kwargs,
        )
        # ESM3 residue scaling on each residual branch.
        hidden_states = residual + attn_output / self.scaling_factor
        residual = hidden_states
        hidden_states = residual + self.mlp(self.post_attention_layernorm(hidden_states)) / self.scaling_factor
        return hidden_states


@auto_docstring
class EsmcPreTrainedModel(PreTrainedModel):
    config_class = EsmcConfig
    base_model_prefix = "esmc"
    supports_gradient_checkpointing = True
    _supports_flash_attn = True
    _supports_sdpa = True
    _supports_flex_attn = True
    _supports_attention_backend = True
    _can_record_outputs = {
        "hidden_states": EsmcLayer,
        "attentions": EsmcAttention,
    }

    # Are kept as non-persistent buffers to avoid being saved in the state dict
    # and causing mismatch when loading from a checkpoint that doesn't have them
    _keys_to_ignore_on_load_unexpected = ["inv_freq", "original_inv_freq"]
    _no_split_modules = ["EsmcLayer"]


@auto_docstring
class EsmcModel(EsmcPreTrainedModel):
    def __init__(self, config: EsmcConfig):
        super().__init__(config)
        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size)
        self.rotary_emb = EsmcRotaryEmbedding(config)
        self.layers = nn.ModuleList([EsmcLayer(config) for _ in range(config.num_hidden_layers)])  # trf-ignore: TRF034
        self.norm = EsmcLayerNorm(config.hidden_size, bias=False)
        self.gradient_checkpointing = False
        self.post_init()

    @merge_with_config_defaults
    @capture_outputs
    @auto_docstring
    def forward(
        self,
        input_ids: torch.Tensor | None = None,
        attention_mask: torch.Tensor | None = None,
        position_ids: torch.Tensor | None = None,
        sequence_id: torch.Tensor | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> BaseModelOutput:
        r"""
        position_ids (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
            Position of each token in the sequence, used by RoPE. Defaults to
            ``arange(sequence_length)``.

            ESMC positions are absolute over the whole input, and keep running *across* chain
            boundaries in a multi-chain input. The chain structure is carried separately by
            ``sequence_id``, which masks attention: unlike the packed-sequence format used elsewhere
            in the library, these are therefore not per-chain position ids restarting at 0, and the
            chain structure is deliberately not inferred from them.
        sequence_id (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
            Integer chain-ID tensor for chain-aware attention masking. Tokens with the same
            non-negative integer value can attend to each other; tokens with different values
            cannot (cross-chain masking). Padding positions should be set to ``-1`` and inputs
            must be **right-padded** (RoPE uses absolute positions starting at 0). When provided,
            ``attention_mask`` is ignored. Passing ``sequence_id`` builds a custom attention
            mask, which requires ``torch>=2.6``. Multi-chain inputs additionally require a
            non-flash ``attn_implementation`` (``'sdpa'`` / ``'eager'`` / ``'flex_attention'``);
            flash attention only supports the single-chain case.

        Examples:

        ```python
        >>> from transformers import AutoTokenizer, EsmcModel

        >>> model = EsmcModel.from_pretrained("biohub/ESMC-300M-hf")
        >>> tokenizer = AutoTokenizer.from_pretrained("biohub/ESMC-300M-hf")
        >>> inputs = tokenizer(["MLKNVQVQLV"], return_tensors="pt")
        >>> outputs = model(**inputs)
        >>> outputs.last_hidden_state.shape
        torch.Size([1, 12, 960])
        ```
        """
        hidden_states = self.embed_tokens(input_ids)
        if position_ids is None:
            position_ids = torch.arange(hidden_states.shape[1], device=hidden_states.device).unsqueeze(0)
        position_embeddings = self.rotary_emb(hidden_states, position_ids)

        if sequence_id is not None:
            if is_flash_attention_requested(self.config):
                raise ValueError(
                    "`sequence_id` (chain-aware attention) is not supported with flash attention, "
                    "which can't represent the block-diagonal chain mask. Re-load the model with "
                    "attn_implementation='sdpa' (or 'eager' / 'flex_attention')."
                )
            attention_mask = create_bidirectional_mask(
                config=self.config,
                inputs_embeds=hidden_states,
                attention_mask=None,
                and_mask_function=packed_sequence_mask_function(sequence_id),
            )
        else:
            attention_mask = create_bidirectional_mask(
                config=self.config,
                inputs_embeds=hidden_states,
                attention_mask=attention_mask,
            )

        for layer in self.layers:
            hidden_states = layer(
                hidden_states,
                attention_mask,
                position_embeddings=position_embeddings,
                **kwargs,
            )

        output = self.norm(hidden_states)

        return BaseModelOutput(last_hidden_state=output)


class EsmcMaskedLMHead(nn.Module):
    def __init__(self, hidden_size: int, vocab_size: int, intermediate_size: int | None = None) -> None:
        super().__init__()
        intermediate_size = intermediate_size if intermediate_size is not None else hidden_size
        self.dense = nn.Linear(hidden_size, intermediate_size)
        self.layer_norm = nn.LayerNorm(intermediate_size)
        self.decoder = nn.Linear(intermediate_size, vocab_size)

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = F.gelu(hidden_states)
        hidden_states = self.layer_norm(hidden_states)
        return self.decoder(hidden_states)


@auto_docstring
class EsmcForMaskedLM(EsmcPreTrainedModel):
    def __init__(self, config: EsmcConfig):
        super().__init__(config)
        self.esmc = EsmcModel(config)
        self.lm_head = EsmcMaskedLMHead(config.hidden_size, config.vocab_size)
        self.post_init()

    # ``lm_head`` is a dense/norm/decoder stack rather than the output projection itself, so the
    # base-class accessors (which return ``lm_head``) would hand ``resize_token_embeddings`` a module
    # with no ``weight``. Same reason ``EsmForMaskedLM`` overrides these.
    def get_output_embeddings(self) -> nn.Linear:
        return self.lm_head.decoder

    def set_output_embeddings(self, new_embeddings: nn.Linear):
        self.lm_head.decoder = new_embeddings

    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: torch.Tensor | None = None,
        attention_mask: torch.Tensor | None = None,
        position_ids: torch.Tensor | None = None,
        sequence_id: torch.Tensor | None = None,
        labels: torch.Tensor | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple[torch.Tensor, ...] | MaskedLMOutput:
        r"""
        sequence_id (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
            Integer chain-ID tensor forwarded to the encoder for chain-aware
            attention masking. See :meth:`EsmcModel.forward` for the encoding.
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for masked language modelling loss.  Positions with label ``-100``
            are ignored.  Other positions must be in ``[0, config.vocab_size)``.

        Examples:

        ```python
        >>> from transformers import AutoTokenizer, EsmcForMaskedLM
        >>> import torch

        >>> model = EsmcForMaskedLM.from_pretrained("biohub/ESMC-300M-hf")
        >>> tokenizer = AutoTokenizer.from_pretrained("biohub/ESMC-300M-hf")
        >>> inputs = tokenizer(["MLKNVQ<mask>LV"], return_tensors="pt")
        >>> outputs = model(**inputs)
        >>> outputs.logits.shape
        torch.Size([1, 11, 64])
        ```
        """
        encoder_outputs = self.esmc(
            input_ids=input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            sequence_id=sequence_id,
            return_dict=True,
            **kwargs,
        )

        logits = self.lm_head(encoder_outputs.last_hidden_state)

        loss: torch.Tensor | None = None
        if labels is not None:
            loss = self.loss_function(logits, labels, vocab_size=self.config.vocab_size, **kwargs)

        return MaskedLMOutput(
            loss=loss,
            logits=logits,
            hidden_states=encoder_outputs.hidden_states,
            attentions=encoder_outputs.attentions,
        )


class EsmcClassificationHead(nn.Module):
    """Head for sentence-level classification tasks."""

    def __init__(self, config: EsmcConfig):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.dropout = nn.Dropout(config.classifier_dropout)
        self.out_proj = nn.Linear(config.hidden_size, config.num_labels)

    def forward(self, features, **kwargs):
        x = features[:, 0, :]  # take <s> token (equiv. to [CLS])
        x = self.dropout(x)
        x = self.dense(x)
        x = torch.tanh(x)
        x = self.dropout(x)
        x = self.out_proj(x)
        return x


@auto_docstring(
    custom_intro="""
    ESMC Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled
    output) e.g. for GLUE tasks.
    """
)
class EsmcForSequenceClassification(EsmcPreTrainedModel):
    def __init__(self, config: EsmcConfig):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.config = config
        self.esmc = EsmcModel(config)
        self.classifier = EsmcClassificationHead(config)

        self.post_init()

    @can_return_tuple
    @auto_docstring
    def forward(
        self,
        input_ids: torch.LongTensor | None = None,
        attention_mask: torch.Tensor | None = None,
        position_ids: torch.LongTensor | None = None,
        inputs_embeds: torch.FloatTensor | None = None,
        labels: torch.LongTensor | None = None,
        **kwargs: Unpack[TransformersKwargs],
    ) -> tuple | SequenceClassifierOutput:
        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).
        """

        outputs = self.esmc(
            input_ids,
            attention_mask=attention_mask,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
            **kwargs,
        )
        sequence_output = outputs[0]
        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            labels = labels.to(logits.device)

            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(logits.squeeze(), labels.squeeze())
                else:
                    loss = loss_fct(logits, labels)
            elif self.config.problem_type == "single_label_classification":
                loss_fct = CrossEntropyLoss()
                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
            elif self.config.problem_type == "multi_label_classification":
                loss_fct = BCEWithLogitsLoss()
                loss = loss_fct(logits, labels)

        return SequenceClassifierOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )


class EsmcForTokenClassification(GenericForTokenClassification, EsmcPreTrainedModel):
    pass


__all__ = [
    "EsmcModel",
    "EsmcForMaskedLM",
    "EsmcForSequenceClassification",
    "EsmcForTokenClassification",
    "EsmcPreTrainedModel",
]