Initial commit

.gitignore

@@ -0,0 +1,5 @@
+data
+out/t5
+out/gpt2
+out/roberta
+.cache_training

bart.py

@@ -0,0 +1,10 @@
+from transformers import BartConfig, BartForSequenceClassification, BartModel
+from torch import nn
+
+class BartForClassification(BartForSequenceClassification):
+    def __init__(self, config: BartConfig):
+        self.config = config
+        self.bart = BartForSequenceClassification(config)
+        self.bart.out_proj = nn.Linear(768, 4)
+        
+

gpt2.py

@@ -0,0 +1,154 @@
+import torch
+from torch import nn
+from transformers import GPT2PreTrainedModel, GPT2Model
+from transformers.modeling_outputs import SequenceClassifierOutputWithPast
+
+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()
+
+
+class GPT2ClassificationHeadCustom(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        hidden_size = config.n_embd
+        self.dense_1_input = nn.Linear(hidden_size, 2 * hidden_size)
+        self.dense_1_hidden = nn.Linear(hidden_size, 2 * hidden_size)
+        self.dense_2 = nn.Linear(4 * hidden_size, 4 * hidden_size)
+        self.dense_3 = nn.Linear(4 * hidden_size, 4 * hidden_size)
+        self.dense_4 = nn.Linear(4 * hidden_size, hidden_size)
+        self.dropout = nn.Dropout(config.resid_pdrop)
+        self.out_proj = nn.Linear(hidden_size, config.num_labels, bias=False)
+
+    def forward(self, x, **kwargs):
+        if 'hidden_states' in kwargs and kwargs['hidden_states'] is not None:
+            # Get hidden states from last layer
+            hidden = kwargs['hidden_states'][-1]
+        else:
+            hidden = torch.zeros(x.size(), dtype=x.dtype, device=x.device)
+
+        x = self.dense_1_input(x)
+        x = torch.relu(x)
+        x = self.dropout(x)
+
+        hidden = self.dense_1_hidden(hidden)
+        hidden = torch.relu(hidden)
+        hidden = self.dropout(hidden)
+
+        x = torch.cat((x, hidden), dim=2)
+        x = self.dense_2(x)
+        x = torch.relu(x)
+        x = self.dense_3(x)
+        x = torch.relu(x)
+        x = self.dense_4(x)
+        x = torch.relu(x)
+        x = self.dropout(x)
+
+        x = self.out_proj(x)
+        return x
+
+class GPT2ForSequenceClassificationCustom(GPT2ForSequenceClassification):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.transformer = GPT2Model(config)
+
+        self.score = GPT2ClassificationHeadCustom(config)
+
+        self.init_weights()
+
+        # Model parallel
+        self.model_parallel = False
+        self.device_map = None
+
+    def forward(
+        self,
+        input_ids=None,
+        past_key_values=None,
+        attention_mask=None,
+        token_type_ids=None,
+        position_ids=None,
+        head_mask=None,
+        inputs_embeds=None,
+        labels=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        r"""
+        labels (:obj:`torch.LongTensor` of shape :obj:`(batch_size,)`, `optional`):
+            Labels for computing the sequence classification/regression loss. Indices should be in :obj:`[0, ...,
+            config.num_labels - 1]`. If :obj:`config.num_labels == 1` a regression loss is computed (Mean-Square loss),
+            If :obj:`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]
+        if return_dict:
+            logits = self.score(hidden_states, hidden_states=transformer_outputs.hidden_states)
+        else:
+            raise NotImplemented('Not implemented for using non-dictionary object')
+
+        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:
+                sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1
+            else:
+                sequence_lengths = -1
+
+        pooled_logits = logits[range(batch_size), sequence_lengths]
+
+        loss = None
+        if labels is not None:
+            if self.num_labels == 1:
+                #  We are doing regression
+                loss_fct = nn.MSELoss()
+                loss = loss_fct(pooled_logits.view(-1), labels.to(self.dtype).view(-1))
+            else:
+                loss_fct = nn.CrossEntropyLoss()
+                loss = loss_fct(pooled_logits.view(-1, self.num_labels), labels.view(-1))
+
+        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,
+        )

out/gpt2_results/README.md

@@ -0,0 +1,53 @@
+---
+tags:
+- generated_from_trainer
+model-index:
+- name: gpt2_results
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# gpt2_results
+
+This model is a fine-tuned version of [out/gpt2](https://huggingface.co/out/gpt2) on an unknown dataset.
+It achieves the following results on the evaluation set:
+- eval_loss: 0.3020
+- eval_accuracy: 0.9195
+- eval_runtime: 24.1139
+- eval_samples_per_second: 82.94
+- eval_steps_per_second: 10.367
+- step: 0
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 2e-05
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- training_steps: 2500
+
+### Framework versions
+
+- Transformers 4.26.1
+- Pytorch 1.13.1+cu117
+- Datasets 2.9.0
+- Tokenizers 0.13.2

out/gpt2_results/all_results.json

@@ -0,0 +1,8 @@
+{
+    "eval_accuracy": 0.9194999933242798,
+    "eval_loss": 0.3020096719264984,
+    "eval_runtime": 24.1139,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 82.94,
+    "eval_steps_per_second": 10.367
+}

out/gpt2_results/eval_results.json

@@ -0,0 +1,8 @@
+{
+    "eval_accuracy": 0.9194999933242798,
+    "eval_loss": 0.3020096719264984,
+    "eval_runtime": 24.1139,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 82.94,
+    "eval_steps_per_second": 10.367
+}

out/roberta_results/README.md

@@ -0,0 +1,53 @@
+---
+tags:
+- generated_from_trainer
+model-index:
+- name: roberta_results
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# roberta_results
+
+This model is a fine-tuned version of [out/roberta](https://huggingface.co/out/roberta) on an unknown dataset.
+It achieves the following results on the evaluation set:
+- eval_loss: 0.2960
+- eval_accuracy: 0.9230
+- eval_runtime: 17.8166
+- eval_samples_per_second: 112.255
+- eval_steps_per_second: 14.032
+- step: 0
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 2e-05
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- training_steps: 2500
+
+### Framework versions
+
+- Transformers 4.26.1
+- Pytorch 1.13.1+cu117
+- Datasets 2.9.0
+- Tokenizers 0.13.2

out/roberta_results/all_results.json

@@ -0,0 +1,8 @@
+{
+    "eval_accuracy": 0.9229999780654907,
+    "eval_loss": 0.29598742723464966,
+    "eval_runtime": 17.8166,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 112.255,
+    "eval_steps_per_second": 14.032
+}

out/roberta_results/eval_results.json

@@ -0,0 +1,8 @@
+{
+    "eval_accuracy": 0.9229999780654907,
+    "eval_loss": 0.29598742723464966,
+    "eval_runtime": 17.8166,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 112.255,
+    "eval_steps_per_second": 14.032
+}

out/t5_results/README.md

@@ -0,0 +1,53 @@
+---
+tags:
+- generated_from_trainer
+model-index:
+- name: t5_results
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# t5_results
+
+This model is a fine-tuned version of [out/t5](https://huggingface.co/out/t5) on an unknown dataset.
+It achieves the following results on the evaluation set:
+- eval_loss: 1.2139
+- eval_accuracy: 0.4675
+- eval_runtime: 40.5651
+- eval_samples_per_second: 49.303
+- eval_steps_per_second: 6.163
+- step: 0
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 2e-05
+- train_batch_size: 8
+- eval_batch_size: 8
+- seed: 42
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: linear
+- training_steps: 2500
+
+### Framework versions
+
+- Transformers 4.26.1
+- Pytorch 1.13.1+cu117
+- Datasets 2.9.0
+- Tokenizers 0.13.2

out/t5_results/all_results.json

@@ -0,0 +1,8 @@
+{
+    "eval_accuracy": 0.4675000011920929,
+    "eval_loss": 1.213880181312561,
+    "eval_runtime": 40.5651,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 49.303,
+    "eval_steps_per_second": 6.163
+}

out/t5_results/eval_results.json

@@ -0,0 +1,8 @@
+{
+    "eval_accuracy": 0.4675000011920929,
+    "eval_loss": 1.213880181312561,
+    "eval_runtime": 40.5651,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 49.303,
+    "eval_steps_per_second": 6.163
+}

preparer_ag_nenws.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+
+import json
+import logging
+from pathlib import Path
+from typing import List, Dict
+
+from datasets import load_dataset
+
+logger = logging.getLogger(__name__)
+
+MAP_LABEL_TRANSLATION = {
+    0: 'world',
+    1: 'sport',
+    2: 'business',
+    3: 'scitech'
+}
+
+
+def save_as_translations(original_save_path: Path, data_to_save: List[Dict]) -> None:
+    file_name = 's2s-' + original_save_path.name
+    file_path = original_save_path.parent / file_name
+
+    print(f'Saving into: {file_path}')
+    with open(file_path, 'wt') as f_write:
+        for data_line in data_to_save:
+            label = data_line['label']
+            new_label = MAP_LABEL_TRANSLATION[label]
+            data_line['label'] = new_label
+            data_line_str = json.dumps(data_line)
+            f_write.write(f'{data_line_str}\n')
+
+
+def main() -> None:
+    loaded_data = load_dataset('ag_news')
+    logger.info(f'Loaded dataset ag_news: {loaded_data}')
+
+    save_path = Path('data/')
+    save_train_path = save_path / 'train.json'
+    save_valid_path = save_path / 'valid.json'
+    save_test_path = save_path / 'test.json'
+    if not save_path.exists():
+        save_path.mkdir()
+
+    # Read train and validation data
+    data_train, data_valid, data_test = [], [], []
+    for source_data, dataset, max_size in [
+        (loaded_data['train'], data_train, None),
+        (loaded_data['test'], data_valid, None)
+    ]:
+        for i, data in enumerate(source_data):
+            if max_size is not None and i >= max_size:
+                break
+            data_line = {
+                'label': int(data['label']),
+                'text': data['text'],
+            }
+            dataset.append(data_line)
+    logger.info(f'Train: {len(data_train):6d}')
+
+    # Split validation set into 2 classes for validation and test splitting
+    world, sport, business, scitech = [], [], [], []
+
+    for data in data_valid:
+        label = data['label']
+        if label == 0:
+            world.append(data)
+        elif label == 1:
+            sport.append(data)
+        elif label == 2:
+            business.append(data)
+        elif label == 3:
+            scitech.append(data)
+
+    logger.info(f'World: {len(world):6d}')
+    logger.info(f'Sport: {len(sport):6d}')
+    logger.info(f'Business: {len(business):6d}')
+    logger.info(f'Scitech: {len(scitech):6d}')
+
+    print(world)
+    print(f'World: {len(world)}')
+    print(f'Sport: {len(sport):6d}')
+    print(f'Business: {len(business):6d}')
+    print(f'Scitech: {len(scitech):6d}')
+
+
+    # Split 2 classes into validation and test
+    size_half_world = int(len(world) / 2)
+    size_half_sport = int(len(sport) / 2)
+    size_half_business = int(len(business) / 2)
+    size_half_scitech = int(len(scitech) / 2)
+    logger.info(f'Valid: {len(data_valid):6d}')
+    logger.info(f'Test : {len(data_test):6d}')
+
+    data_valid = world[:size_half_world] + sport[:size_half_sport] + business[:size_half_business] + scitech[:size_half_scitech]
+    data_test = world[size_half_world:] + sport[size_half_sport:] + business[size_half_business:] + scitech[size_half_scitech:]
+
+    # Save files
+    for file_path, data_to_save in [
+        (save_train_path, data_train),
+        (save_valid_path, data_valid),
+        (save_test_path, data_test)
+    ]:
+        print(f'Saving into: {file_path}')
+        with open(file_path, 'wt') as f_write:
+            for data_line in data_to_save:
+                data_line_str = json.dumps(data_line)
+                f_write.write(f'{data_line_str}\n')
+
+        save_as_translations(file_path, data_to_save)
+
+
+if __name__ == '__main__':
+    main()

roberta.py

@@ -0,0 +1,54 @@
+from typing import Optional, Union, Tuple
+
+import torch
+from torch import nn
+from torch.nn import MSELoss, CrossEntropyLoss, BCEWithLogitsLoss
+from transformers import RobertaForSequenceClassification, RobertaModel
+from transformers.modeling_outputs import SequenceClassifierOutput
+
+
+# Simple version #
+
+class RobertaClassificationHeadCustomSimple(nn.Module):
+    """Head for sentence-level classification tasks."""
+
+    def __init__(self, config):
+        super().__init__()
+        hidden_size = config.hidden_size
+        self.dense_1 = nn.Linear(hidden_size, 4 * hidden_size)
+        self.dense_2 = nn.Linear(4 * hidden_size, hidden_size)
+        classifier_dropout = (
+            config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
+        )
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.out_proj = nn.Linear(hidden_size, config.num_labels)
+        self.activation = nn.GELU()
+
+    def forward(self, features, **kwargs):
+        x = features[:, 0, :]  # take <s> token (equiv. to [CLS])
+
+        x = self.dense_1(x)
+        x = self.activation(x)
+        x = self.dropout(x)
+
+        x = self.dense_2(x)
+        x = self.activation(x)
+        x = self.dropout(x)
+
+        x = self.out_proj(x)
+        return x
+
+
+class RobertaForSequenceClassificationCustomSimple(RobertaForSequenceClassification):
+    _keys_to_ignore_on_load_missing = [r"position_ids"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.config = config
+
+        self.roberta = RobertaModel(config, add_pooling_layer=False)
+        self.classifier = RobertaClassificationHeadCustomSimple(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()

run_glue.py

@@ -0,0 +1,685 @@
+#!/usr/bin/env python
+# coding=utf-8
+# Copyright 2020 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.
+""" Finetuning the library models for sequence classification on GLUE."""
+# You can also adapt this script on your own text classification task. Pointers for this are left as comments.
+
+import logging
+import os
+import random
+import sys
+from collections import defaultdict
+from dataclasses import dataclass, field
+from typing import Optional
+
+import datasets
+import numpy as np
+from datasets import load_dataset
+
+import evaluate
+import transformers
+from transformers import (
+    AutoConfig,
+    AutoModelForSequenceClassification,
+    AutoTokenizer,
+    DataCollatorWithPadding,
+    EvalPrediction,
+    HfArgumentParser,
+    PretrainedConfig,
+    Trainer,
+    TrainingArguments,
+    default_data_collator,
+    set_seed,
+)
+from transformers.trainer_utils import get_last_checkpoint
+from transformers.utils import check_min_version, send_example_telemetry
+from transformers.utils.versions import require_version
+
+from roberta import RobertaForSequenceClassificationCustomSimple
+from gpt2 import GPT2ForSequenceClassificationCustom
+from t5 import T5ForClassification
+from transformers import BartForSequenceClassification
+
+MODEL_NAME_TO_CLASS = {
+    'roberta_simple': RobertaForSequenceClassificationCustomSimple,
+    'gpt2_hidden': GPT2ForSequenceClassificationCustom,
+    't5_custom': T5ForClassification,
+    'bart_base': BartForSequenceClassification,
+}
+
+# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
+check_min_version("4.23.0")
+
+require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")
+
+task_to_keys = {
+    "cola": ("sentence", None),
+    "mnli": ("premise", "hypothesis"),
+    "mrpc": ("sentence1", "sentence2"),
+    "qnli": ("question", "sentence"),
+    "qqp": ("question1", "question2"),
+    "rte": ("sentence1", "sentence2"),
+    "sst2": ("sentence", None),
+    "stsb": ("sentence1", "sentence2"),
+    "wnli": ("sentence1", "sentence2"),
+}
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class DataTrainingArguments:
+    """
+    Arguments pertaining to what data we are going to input our model for training and eval.
+
+    Using `HfArgumentParser` we can turn this class
+    into argparse arguments to be able to specify them on
+    the command line.
+    """
+
+    task_name: Optional[str] = field(
+        default=None,
+        metadata={"help": "The name of the task to train on: " + ", ".join(task_to_keys.keys())},
+    )
+    dataset_name: Optional[str] = field(
+        default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
+    )
+    dataset_config_name: Optional[str] = field(
+        default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
+    )
+    max_seq_length: int = field(
+        default=128,
+        metadata={
+            "help": (
+                "The maximum total input sequence length after tokenization. Sequences longer "
+                "than this will be truncated, sequences shorter will be padded."
+            )
+        },
+    )
+    overwrite_cache: bool = field(
+        default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
+    )
+    pad_to_max_length: bool = field(
+        default=True,
+        metadata={
+            "help": (
+                "Whether to pad all samples to `max_seq_length`. "
+                "If False, will pad the samples dynamically when batching to the maximum length in the batch."
+            )
+        },
+    )
+    max_train_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of training examples to this "
+                "value if set."
+            )
+        },
+    )
+    max_eval_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of evaluation examples to this "
+                "value if set."
+            )
+        },
+    )
+    max_predict_samples: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "For debugging purposes or quicker training, truncate the number of prediction examples to this "
+                "value if set."
+            )
+        },
+    )
+    train_file: Optional[str] = field(
+        default=None, metadata={"help": "A csv or a json file containing the training data."}
+    )
+    validation_file: Optional[str] = field(
+        default=None, metadata={"help": "A csv or a json file containing the validation data."}
+    )
+    test_file: Optional[str] = field(default=None, metadata={"help": "A csv or a json file containing the test data."})
+
+    def __post_init__(self):
+        if self.task_name is not None:
+            self.task_name = self.task_name.lower()
+            if self.task_name not in task_to_keys.keys():
+                raise ValueError("Unknown task, you should pick one in " + ",".join(task_to_keys.keys()))
+        elif self.dataset_name is not None:
+            pass
+        elif self.train_file is None or self.validation_file is None:
+            raise ValueError("Need either a GLUE task, a training/validation file or a dataset name.")
+        else:
+            train_extension = self.train_file.split(".")[-1]
+            assert train_extension in ["csv", "json"], "`train_file` should be a csv or a json file."
+            validation_extension = self.validation_file.split(".")[-1]
+            assert (
+                validation_extension == train_extension
+            ), "`validation_file` should have the same extension (csv or json) as `train_file`."
+
+
+@dataclass
+class ModelArguments:
+    """
+    Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
+    """
+
+    model_name_or_path: str = field(
+        metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
+    )
+    config_name: Optional[str] = field(
+        default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
+    )
+    tokenizer_name: Optional[str] = field(
+        default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
+    )
+    cache_dir: Optional[str] = field(
+        default=None,
+        metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
+    )
+    use_fast_tokenizer: bool = field(
+        default=True,
+        metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
+    )
+    model_revision: str = field(
+        default="main",
+        metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
+    )
+    use_auth_token: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "Will use the token generated when running `huggingface-cli login` (necessary to use this script "
+                "with private models)."
+            )
+        },
+    )
+    ignore_mismatched_sizes: bool = field(
+        default=False,
+        metadata={"help": "Will enable to load a pretrained model whose head dimensions are different."},
+    )
+    custom_model: str = field(
+        default=None,
+        metadata={
+            "help": "Use custom implementation from available list",
+            "choices": list(MODEL_NAME_TO_CLASS.keys()),
+        },
+    )
+
+
+def main():
+    # See all possible arguments in src/transformers/training_args.py
+    # or by passing the --help flag to this script.
+    # We now keep distinct sets of args, for a cleaner separation of concerns.
+
+    parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
+    if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
+        # If we pass only one argument to the script and it's the path to a json file,
+        # let's parse it to get our arguments.
+        model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
+    else:
+        model_args, data_args, training_args = parser.parse_args_into_dataclasses()
+
+    # Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
+    # information sent is the one passed as arguments along with your Python/PyTorch versions.
+    send_example_telemetry("run_glue", model_args, data_args)
+
+    if 'bart' in model_args.model_name_or_path:
+        model_args.ignore_mismatched_sizes = True
+    
+    # Setup logging
+    logging.basicConfig(
+        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        handlers=[logging.StreamHandler(sys.stdout)],
+    )
+
+    log_level = training_args.get_process_log_level()
+    logger.setLevel(log_level)
+    datasets.utils.logging.set_verbosity(log_level)
+    transformers.utils.logging.set_verbosity(log_level)
+    transformers.utils.logging.enable_default_handler()
+    transformers.utils.logging.enable_explicit_format()
+
+    # Log on each process the small summary:
+    logger.warning(
+        f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+        + f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
+    )
+    logger.info(f"Training/evaluation parameters {training_args}")
+
+    # Detecting last checkpoint.
+    last_checkpoint = None
+    if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
+        last_checkpoint = get_last_checkpoint(training_args.output_dir)
+        if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
+            raise ValueError(
+                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
+                "Use --overwrite_output_dir to overcome."
+            )
+        elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
+            logger.info(
+                f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
+                "the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
+            )
+
+    # Set seed before initializing model.
+    set_seed(training_args.seed)
+
+    # Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
+    # or specify a GLUE benchmark task (the dataset will be downloaded automatically from the datasets Hub).
+    #
+    # For CSV/JSON files, this script will use as labels the column called 'label' and as pair of sentences the
+    # sentences in columns called 'sentence1' and 'sentence2' if such column exists or the first two columns not named
+    # label if at least two columns are provided.
+    #
+    # If the CSVs/JSONs contain only one non-label column, the script does single sentence classification on this
+    # single column. You can easily tweak this behavior (see below)
+    #
+    # In distributed training, the load_dataset function guarantee that only one local process can concurrently
+    # download the dataset.
+    if data_args.task_name is not None:
+        # Downloading and loading a dataset from the hub.
+        raw_datasets = load_dataset(
+            "glue",
+            data_args.task_name,
+            cache_dir=model_args.cache_dir,
+            use_auth_token=True if model_args.use_auth_token else None,
+        )
+    elif data_args.dataset_name is not None:
+        # Downloading and loading a dataset from the hub.
+        raw_datasets = load_dataset(
+            data_args.dataset_name,
+            data_args.dataset_config_name,
+            cache_dir=model_args.cache_dir,
+            use_auth_token=True if model_args.use_auth_token else None,
+        )
+    else:
+        # Loading a dataset from your local files.
+        # CSV/JSON training and evaluation files are needed.
+        data_files = {"train": data_args.train_file, "validation": data_args.validation_file}
+
+        # Get the test dataset: you can provide your own CSV/JSON test file (see below)
+        # when you use `do_predict` without specifying a GLUE benchmark task.
+        if training_args.do_predict:
+            if data_args.test_file is not None:
+                train_extension = data_args.train_file.split(".")[-1]
+                test_extension = data_args.test_file.split(".")[-1]
+                assert (
+                    test_extension == train_extension
+                ), "`test_file` should have the same extension (csv or json) as `train_file`."
+                data_files["test"] = data_args.test_file
+            else:
+                raise ValueError("Need either a GLUE task or a test file for `do_predict`.")
+
+        for key in data_files.keys():
+            logger.info(f"load a local file for {key}: {data_files[key]}")
+
+        if data_args.train_file.endswith(".csv"):
+            # Loading a dataset from local csv files
+            raw_datasets = load_dataset(
+                "csv",
+                data_files=data_files,
+                cache_dir=model_args.cache_dir,
+                use_auth_token=True if model_args.use_auth_token else None,
+            )
+        else:
+            # Loading a dataset from local json files
+            raw_datasets = load_dataset(
+                "json",
+                data_files=data_files,
+                cache_dir=model_args.cache_dir,
+                use_auth_token=True if model_args.use_auth_token else None,
+            )
+    # See more about loading any type of standard or custom dataset at
+    # https://huggingface.co/docs/datasets/loading_datasets.html.
+
+    # Labels
+    if data_args.task_name is not None:
+        is_regression = data_args.task_name == "stsb"
+        if not is_regression:
+            label_list = raw_datasets["train"].features["label"].names
+            num_labels = len(label_list)
+        else:
+            num_labels = 1
+    else:
+        # Trying to have good defaults here, don't hesitate to tweak to your needs.
+        is_regression = raw_datasets["train"].features["label"].dtype in ["float32", "float64"]
+        if is_regression:
+            num_labels = 1
+        else:
+            # A useful fast method:
+            # https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.unique
+            label_list = raw_datasets["train"].unique("label")
+            label_list.sort()  # Let's sort it for determinism
+            num_labels = len(label_list)
+
+    # Load pretrained model and tokenizer
+    #
+    # In distributed training, the .from_pretrained methods guarantee that only one local process can concurrently
+    # download model & vocab.
+    config = AutoConfig.from_pretrained(
+        model_args.config_name if model_args.config_name else model_args.model_name_or_path,
+        num_labels=num_labels,
+        finetuning_task=data_args.task_name,
+        cache_dir=model_args.cache_dir,
+        revision=model_args.model_revision,
+        use_auth_token=True if model_args.use_auth_token else None,
+    )
+    tokenizer = AutoTokenizer.from_pretrained(
+        model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
+        cache_dir=model_args.cache_dir,
+        use_fast=model_args.use_fast_tokenizer,
+        revision=model_args.model_revision,
+        use_auth_token=True if model_args.use_auth_token else None,
+    )
+    custom_model = model_args.custom_model
+
+    if custom_model is not None:
+        # Check model and implementation is the same
+        if 'roberta' in custom_model and 'roberta' not in model_args.model_name_or_path:
+            raise RuntimeError('Model and custom implementation should be the same type: RoBERTa')
+        elif 'gpt2' in custom_model and 'gpt2' not in model_args.model_name_or_path:
+            raise RuntimeError('Model and custom implementation should be the same type: GPT-2')
+
+        # Set custom configuration in model configuration
+        config.use_hidden_states = 'hidden' in custom_model
+        logger.info(f'Using hidden states in model: {config.use_hidden_states}')
+
+        print(f'-------------------------------------------------------- Using hidden: {config.use_hidden_states}')
+
+        # Get class to initialize model
+        model_cls = MODEL_NAME_TO_CLASS[custom_model]
+    else:
+        model_cls = AutoModelForSequenceClassification
+    logger.info(f'Using implementation from class: {model_cls.__name__}')
+    model = model_cls.from_pretrained(
+        model_args.model_name_or_path,
+        from_tf=bool(".ckpt" in model_args.model_name_or_path),
+        config=config,
+        cache_dir=model_args.cache_dir,
+        revision=model_args.model_revision,
+        use_auth_token=True if model_args.use_auth_token else None,
+        ignore_mismatched_sizes=model_args.ignore_mismatched_sizes,        
+    )
+
+    print(model)
+
+    if 'gpt2' in tokenizer.name_or_path and tokenizer.pad_token is None:
+        logger.info(f'Set PAD token to EOS: {tokenizer.eos_token}')
+        tokenizer._pad_token = tokenizer.eos_token
+        model.config.pad_token_id = model.config.eos_token_id
+
+    # Preprocessing the raw_datasets
+    if data_args.task_name is not None:
+         sentence1_key, sentence2_key = task_to_keys[data_args.task_name]
+
+    # Preprocessing the raw_datasets
+    if data_args.task_name is not None:
+        sentence1_key, sentence2_key = task_to_keys[data_args.task_name]
+    else:
+        # Again, we try to have some nice defaults but don't hesitate to tweak to your use case.
+        non_label_column_names = [name for name in raw_datasets["train"].column_names if name != "label"]
+        if "sentence1" in non_label_column_names and "sentence2" in non_label_column_names:
+            sentence1_key, sentence2_key = "sentence1", "sentence2"
+        else:
+            if len(non_label_column_names) >= 2:
+                sentence1_key, sentence2_key = non_label_column_names[:2]
+            else:
+                sentence1_key, sentence2_key = non_label_column_names[0], None
+
+    # Padding strategy
+    if data_args.pad_to_max_length:
+        padding = "max_length"
+    else:
+        # We will pad later, dynamically at batch creation, to the max sequence length in each batch
+        padding = False
+
+    # Some models have set the order of the labels to use, so let's make sure we do use it.
+    label_to_id = None
+    if (
+        model.config.label2id != PretrainedConfig(num_labels=num_labels).label2id
+        and data_args.task_name is not None
+        and not is_regression
+    ):
+        # Some have all caps in their config, some don't.
+        label_name_to_id = {k.lower(): v for k, v in model.config.label2id.items()}
+        if list(sorted(label_name_to_id.keys())) == list(sorted(label_list)):
+            label_to_id = {i: int(label_name_to_id[label_list[i]]) for i in range(num_labels)}
+        else:
+            logger.warning(
+                "Your model seems to have been trained with labels, but they don't match the dataset: ",
+                f"model labels: {list(sorted(label_name_to_id.keys()))}, dataset labels: {list(sorted(label_list))}."
+                "\nIgnoring the model labels as a result.",
+            )
+    elif data_args.task_name is None and not is_regression:
+        label_to_id = {v: i for i, v in enumerate(label_list)}
+
+    if label_to_id is not None:
+        model.config.label2id = label_to_id
+        model.config.id2label = {id: label for label, id in config.label2id.items()}
+    elif data_args.task_name is not None and not is_regression:
+        model.config.label2id = {l: i for i, l in enumerate(label_list)}
+        model.config.id2label = {id: label for label, id in config.label2id.items()}
+
+    if data_args.max_seq_length > tokenizer.model_max_length:
+        logger.warning(
+            f"The max_seq_length passed ({data_args.max_seq_length}) is larger than the maximum length for the"
+            f"model ({tokenizer.model_max_length}). Using max_seq_length={tokenizer.model_max_length}."
+        )
+    max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)
+
+    def preprocess_function(examples):
+        # Tokenize the texts
+        args = (
+            (examples[sentence1_key],) if sentence2_key is None else (examples[sentence1_key], examples[sentence2_key])
+        )
+        result = tokenizer(*args, padding=padding, max_length=max_seq_length, truncation=True)
+
+        # Map labels to IDs (not necessary for GLUE tasks)
+        if label_to_id is not None and "label" in examples:
+            result["label"] = [(label_to_id[l] if l != -1 else -1) for l in examples["label"]]
+        return result
+
+    with training_args.main_process_first(desc="dataset map pre-processing"):
+        raw_datasets = raw_datasets.map(
+            preprocess_function,
+            batched=True,
+            load_from_cache_file=not data_args.overwrite_cache,
+            desc="Running tokenizer on dataset",
+        )
+    if training_args.do_train:
+        if "train" not in raw_datasets:
+            raise ValueError("--do_train requires a train dataset")
+        train_dataset = raw_datasets["train"]
+        if data_args.max_train_samples is not None:
+            max_train_samples = min(len(train_dataset), data_args.max_train_samples)
+            train_dataset = train_dataset.select(range(max_train_samples))
+
+    if training_args.do_eval:
+        if "validation" not in raw_datasets and "validation_matched" not in raw_datasets:
+            raise ValueError("--do_eval requires a validation dataset")
+        eval_dataset = raw_datasets["validation_matched" if data_args.task_name == "mnli" else "validation"]
+        if data_args.max_eval_samples is not None:
+            max_eval_samples = min(len(eval_dataset), data_args.max_eval_samples)
+            label_to_indexes = defaultdict(list)
+            for index, eval_sample in enumerate(eval_dataset):
+                label_to_indexes[eval_sample['label']].append(index)
+            max_samples_per_label = int(max_eval_samples / len(label_to_indexes))
+            eval_sample_indexes = []
+            for label, indexes in label_to_indexes.items():
+                eval_sample_indexes.extend(indexes[:max_samples_per_label])
+                logger.info(f"Set {max_samples_per_label} samples for {label}-class")
+            eval_sample_indexes.sort()
+            eval_dataset = eval_dataset.select(eval_sample_indexes)
+
+    if training_args.do_predict or data_args.task_name is not None or data_args.test_file is not None:
+        if "test" not in raw_datasets and "test_matched" not in raw_datasets:
+            raise ValueError("--do_predict requires a test dataset")
+        predict_dataset = raw_datasets["test_matched" if data_args.task_name == "mnli" else "test"]
+        if data_args.max_predict_samples is not None:
+            max_predict_samples = min(len(predict_dataset), data_args.max_predict_samples)
+            predict_dataset = predict_dataset.select(range(max_predict_samples))
+
+    # Log a few random samples from the training set:
+    if training_args.do_train:
+        for index in random.sample(range(len(train_dataset)), 3):
+            logger.info(f"Sample {index} of the training set: {train_dataset[index]}.")
+
+    # Get the metric function
+    if data_args.task_name is not None:
+        metric = evaluate.load("glue", data_args.task_name)
+    else:
+        metric = evaluate.load("accuracy")
+
+    # You can define your custom compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with a
+    # predictions and label_ids field) and has to return a dictionary string to float.
+    def compute_metrics(p: EvalPrediction):
+        preds = p.predictions[0] if isinstance(p.predictions, tuple) else p.predictions
+        preds = np.squeeze(preds) if is_regression else np.argmax(preds, axis=1)
+        if data_args.task_name is not None:
+            result = metric.compute(predictions=preds, references=p.label_ids)
+            if len(result) > 1:
+                result["combined_score"] = np.mean(list(result.values())).item()
+            return result
+        elif is_regression:
+            return {"mse": ((preds - p.label_ids) ** 2).mean().item()}
+        else:
+            return {"accuracy": (preds == p.label_ids).astype(np.float32).mean().item()}
+
+    # Data collator will default to DataCollatorWithPadding when the tokenizer is passed to Trainer, so we change it if
+    # we already did the padding.
+    if data_args.pad_to_max_length:
+        data_collator = default_data_collator
+    elif training_args.fp16:
+        data_collator = DataCollatorWithPadding(tokenizer, pad_to_multiple_of=8)
+    else:
+        data_collator = None
+
+    # Initialize our Trainer
+    trainer = Trainer(
+        model=model,
+        args=training_args,
+        train_dataset=train_dataset if training_args.do_train else None,
+        eval_dataset=eval_dataset if training_args.do_eval else None,
+        compute_metrics=compute_metrics,
+        tokenizer=tokenizer,
+        data_collator=data_collator,
+    )
+
+    # Training
+    ignore_keys_for_eval = ['hidden_states', 'attentions', 'past_key_values']
+    if training_args.do_train:
+        checkpoint = None
+        if training_args.resume_from_checkpoint is not None:
+            checkpoint = training_args.resume_from_checkpoint
+        elif last_checkpoint is not None:
+            checkpoint = last_checkpoint
+        train_result = trainer.train(resume_from_checkpoint=checkpoint, ignore_keys_for_eval=ignore_keys_for_eval)
+        metrics = train_result.metrics
+        max_train_samples = (
+            data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset)
+        )
+        metrics["train_samples"] = min(max_train_samples, len(train_dataset))
+
+        trainer.save_model()  # Saves the tokenizer too for easy upload
+
+        trainer.log_metrics("train", metrics)
+        trainer.save_metrics("train", metrics)
+        trainer.save_state()
+
+    # Evaluation
+    if training_args.do_eval:
+        logger.info("*** Evaluate ***")
+
+        # Loop to handle MNLI double evaluation (matched, mis-matched)
+        tasks = [data_args.task_name]
+        eval_datasets = [eval_dataset]
+        if data_args.task_name == "mnli":
+            tasks.append("mnli-mm")
+            valid_mm_dataset = raw_datasets["validation_mismatched"]
+            if data_args.max_eval_samples is not None:
+                max_eval_samples = min(len(valid_mm_dataset), data_args.max_eval_samples)
+                valid_mm_dataset = valid_mm_dataset.select(range(max_eval_samples))
+            eval_datasets.append(valid_mm_dataset)
+            combined = {}
+
+        for eval_dataset, task in zip(eval_datasets, tasks):
+            metrics = trainer.evaluate(eval_dataset=eval_dataset, ignore_keys=ignore_keys_for_eval)
+
+            max_eval_samples = (
+                data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
+            )
+            metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
+
+            if task == "mnli-mm":
+                metrics = {k + "_mm": v for k, v in metrics.items()}
+            if task is not None and "mnli" in task:
+                combined.update(metrics)
+
+            trainer.log_metrics("eval", metrics)
+            trainer.save_metrics("eval", combined if task is not None and "mnli" in task else metrics)
+
+    if training_args.do_predict:
+        logger.info("*** Predict ***")
+
+        # Loop to handle MNLI double evaluation (matched, mis-matched)
+        tasks = [data_args.task_name]
+        predict_datasets = [predict_dataset]
+        if data_args.task_name == "mnli":
+            tasks.append("mnli-mm")
+            predict_datasets.append(raw_datasets["test_mismatched"])
+
+        for predict_dataset, task in zip(predict_datasets, tasks):
+            # Removing the `label` columns because it contains -1 and Trainer won't like that.
+            predict_dataset = predict_dataset.remove_columns("label")
+            predictions = trainer.predict(predict_dataset, metric_key_prefix="predict", ignore_keys=ignore_keys_for_eval).predictions
+            predictions = np.squeeze(predictions) if is_regression else np.argmax(predictions, axis=1)
+
+            output_predict_file = os.path.join(training_args.output_dir, f"predict_results_{task}.txt")
+            if trainer.is_world_process_zero():
+                with open(output_predict_file, "w") as writer:
+                    logger.info(f"***** Predict results {task} *****")
+                    writer.write("index\tprediction\n")
+                    for index, item in enumerate(predictions):
+                        if is_regression:
+                            writer.write(f"{index}\t{item:3.3f}\n")
+                        else:
+                            item = label_list[item]
+                            writer.write(f"{index}\t{item}\n")
+
+    kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "text-classification"}
+    if data_args.task_name is not None:
+        kwargs["language"] = "en"
+        kwargs["dataset_tags"] = "glue"
+        kwargs["dataset_args"] = data_args.task_name
+        kwargs["dataset"] = f"GLUE {data_args.task_name.upper()}"
+
+    if training_args.push_to_hub:
+        trainer.push_to_hub(**kwargs)
+    else:
+        trainer.create_model_card(**kwargs)
+
+
+def _mp_fn(index):
+    # For xla_spawn (TPUs)
+    main()
+
+
+if __name__ == "__main__":
+    main()

t5.py

@@ -0,0 +1,125 @@
+import torch
+import copy
+from torch import nn
+from transformers import T5PreTrainedModel, T5Config
+from transformers.models.t5.modeling_t5 import T5Stack
+from transformers.modeling_outputs import SequenceClassifierOutput
+
+
+class T5ClassificationHead(nn.Module):
+    def __init__(self, config: T5Config):
+        super().__init__()
+
+        self.dense_in = nn.Linear(config.d_model, 768)
+        self.dense = nn.Linear(768, 768)
+        self.dense_out = nn.Linear(768, config.num_labels)
+        self.dropout = nn.Dropout(0.1)
+
+    def forward(self, features, **kwargs):
+        x = features[:, 0, :]
+        x = self.dropout(x)
+        x = self.dense_in(x)
+        x = torch.relu(x)
+        x = self.dropout(x)
+        x = self.dense(x)
+        x = torch.relu(x)
+        x = self.dropout(x)
+        x = self.dense_out(x)
+
+        return x
+
+
+class T5ForClassification(T5PreTrainedModel):
+    def __init__(self, config: T5Config):
+        super().__init__(config)
+        self.model_dim = config.d_model
+
+        self.shared = nn.Embedding(config.vocab_size, config.d_model)
+
+        encoder_config = copy.deepcopy(config)
+        encoder_config.is_decoder = False
+        encoder_config.use_cache = False
+        encoder_config.is_encoder_decoder = False
+        self.encoder = T5Stack(encoder_config, self.shared)
+
+        decoder_config = copy.deepcopy(config)
+        decoder_config.is_decoder = True
+        decoder_config.is_encoder_decoder = False
+        decoder_config.num_layers = config.num_decoder_layers
+        self.decoder = T5Stack(decoder_config, self.shared)
+
+        modules_to_freeze = [self.encoder.block[i].layer[0] for i in range(len(self.encoder.block))]
+        modules_to_freeze.extend([self.decoder.block[i].layer[0] for i in range(len(self.decoder.block))])
+        modules_to_freeze.extend([self.decoder.block[i].layer[1] for i in range(len(self.decoder.block))])
+
+        for module in modules_to_freeze:
+            for param in module.parameters():
+                param.requires_grad = False
+
+        self.lm_head = T5ClassificationHead(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+        # Model parallel
+        self.model_parallel = False
+        self.device_map = None
+
+
+    def forward(
+            self,
+            input_ids=None,
+            attention_mask=None,
+            head_mask=None,
+            cross_attn_head_mask=None,
+            past_key_values=None,
+            inputs_embeds=None,
+            decoder_inputs_embeds=None,
+            use_cache=None,
+            output_attentions=None,
+            output_hidden_states=None,
+            return_dict=None,
+            labels=None
+            ):
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.encoder(
+                input_ids,
+                attention_mask=attention_mask,
+                head_mask=head_mask,
+                cross_attn_head_mask=cross_attn_head_mask,
+                past_key_values=past_key_values,
+                inputs_embeds=inputs_embeds,
+                use_cache=use_cache,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            )
+
+        outputs = self.decoder(
+                input_ids,
+                attention_mask=attention_mask,
+                head_mask=head_mask,
+                cross_attn_head_mask=cross_attn_head_mask,
+                past_key_values=past_key_values,
+                inputs_embeds=inputs_embeds,
+                use_cache=use_cache,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            )
+
+
+        logits = self.lm_head(outputs[0])
+
+
+        loss = None
+        if labels is not None:
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(logits.view(-1, self.config.num_labels), labels.view(-1))
+
+
+        return SequenceClassifierOutput(
+            loss=loss,
+            logits=logits,
+        )