Albert pretrain datasets/ datacollator (#6168)

* add dataset for albert pretrain

* datacollator for albert pretrain

* naming, comprehension, file reading change

* data cleaning is no needed after this modification

* delete prints

* fix a bug

* file structure change

* add tests for albert datacollator

* remove random seed

* add back len and get item function

* sample file for testing and test code added

* format change for black

* more format change

* Style

* var assignment issue resolve

* add back wrongly deleted DataCollatorWithPadding in init file

* Style
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>

src/transformers/__init__.py

@@ -207,6 +207,7 @@ if is_torch_available():
         DataCollatorForLanguageModeling,
         DataCollatorForNextSentencePrediction,
         DataCollatorForPermutationLanguageModeling,
+        DataCollatorForSOP,
         DataCollatorWithPadding,
         default_data_collator,
     )
@@ -214,6 +215,7 @@ if is_torch_available():
         GlueDataset,
         GlueDataTrainingArguments,
         LineByLineTextDataset,
+        LineByLineWithSOPTextDataset,
         SquadDataset,
         SquadDataTrainingArguments,
         TextDataset,

src/transformers/data/data_collator.py

@@ -198,6 +198,75 @@ class DataCollatorForLanguageModeling:
         return inputs, labels
 
 
+@dataclass
+class DataCollatorForSOP(DataCollatorForLanguageModeling):
+    """
+    Data collator used for sentence order prediction task.
+    - collates batches of tensors, honoring their tokenizer's pad_token
+    - preprocesses batches for both masked language modeling and sentence order prediction
+    """
+
+    def __call__(self, examples: List[Dict[str, torch.Tensor]]) -> Dict[str, torch.Tensor]:
+        input_ids = [example["input_ids"] for example in examples]
+        input_ids = self._tensorize_batch(input_ids)
+        input_ids, labels, attention_mask = self.mask_tokens(input_ids)
+
+        token_type_ids = [example["token_type_ids"] for example in examples]
+        # size of segment_ids varied because randomness, padding zero to the end as the orignal implementation
+        token_type_ids = pad_sequence(token_type_ids, batch_first=True, padding_value=self.tokenizer.pad_token_id)
+
+        sop_label_list = [example["sentence_order_label"] for example in examples]
+        sentence_order_label = torch.stack(sop_label_list)
+
+        return {
+            "input_ids": input_ids,
+            "labels": labels,
+            "attention_mask": attention_mask,
+            "token_type_ids": token_type_ids,
+            "sentence_order_label": sentence_order_label,
+        }
+
+    def mask_tokens(self, inputs: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        Prepare masked tokens inputs/labels/attention_mask for masked language modeling: 80% MASK, 10% random, 10% original.
+        N-gram not applied yet.
+        """
+        if self.tokenizer.mask_token is None:
+            raise ValueError(
+                "This tokenizer does not have a mask token which is necessary for masked language modeling. Remove the --mlm flag if you want to use this tokenizer."
+            )
+
+        labels = inputs.clone()
+        # We sample a few tokens in each sequence for masked-LM training (with probability args.mlm_probability defaults to 0.15 in Bert/RoBERTa)
+        probability_matrix = torch.full(labels.shape, self.mlm_probability)
+        special_tokens_mask = [
+            self.tokenizer.get_special_tokens_mask(val, already_has_special_tokens=True) for val in labels.tolist()
+        ]
+        probability_matrix.masked_fill_(torch.tensor(special_tokens_mask, dtype=torch.bool), value=0.0)
+        if self.tokenizer._pad_token is not None:
+            padding_mask = labels.eq(self.tokenizer.pad_token_id)
+            probability_matrix.masked_fill_(padding_mask, value=0.0)
+        masked_indices = torch.bernoulli(probability_matrix).bool()
+        # probability be `1` (masked), however in albert model attention mask `0` means masked, revert the value
+        attention_mask = (~masked_indices).float()
+        if self.tokenizer._pad_token is not None:
+            attention_padding_mask = labels.eq(self.tokenizer.pad_token_id)
+            attention_mask.masked_fill_(attention_padding_mask, value=1.0)
+        labels[~masked_indices] = -100  # We only compute loss on masked tokens, -100 is default for CE compute
+
+        # 80% of the time, we replace masked input tokens with tokenizer.mask_token ([MASK])
+        indices_replaced = torch.bernoulli(torch.full(labels.shape, 0.8)).bool() & masked_indices
+        inputs[indices_replaced] = self.tokenizer.convert_tokens_to_ids(self.tokenizer.mask_token)
+
+        # 10% of the time, we replace masked input tokens with random word
+        indices_random = torch.bernoulli(torch.full(labels.shape, 0.5)).bool() & masked_indices & ~indices_replaced
+        random_words = torch.randint(len(self.tokenizer), labels.shape, dtype=torch.long)
+        inputs[indices_random] = random_words[indices_random]
+
+        # The rest of the time (10% of the time) we keep the masked input tokens unchanged
+        return inputs, labels, attention_mask
+
+
 @dataclass
 class DataCollatorForPermutationLanguageModeling:
     """

src/transformers/data/datasets/__init__.py

@@ -3,5 +3,10 @@
 # module, but to preserve other warnings. So, don't check this module at all.
 
 from .glue import GlueDataset, GlueDataTrainingArguments
-from .language_modeling import LineByLineTextDataset, TextDataset, TextDatasetForNextSentencePrediction
+from .language_modeling import (
+    LineByLineTextDataset,
+    LineByLineWithSOPTextDataset,
+    TextDataset,
+    TextDatasetForNextSentencePrediction,
+)
 from .squad import SquadDataset, SquadDataTrainingArguments

src/transformers/data/datasets/language_modeling.py

 import os
 import pickle
+import random
 import time
-from typing import Optional
+from typing import Dict, Optional
 
 import torch
 from torch.utils.data.dataset import Dataset
@@ -113,6 +114,147 @@ class LineByLineTextDataset(Dataset):
         return torch.tensor(self.examples[i], dtype=torch.long)
 
 
+class LineByLineWithSOPTextDataset(Dataset):
+    """
+    Dataset for sentence order prediction task, prepare sentence pairs for SOP task
+    """
+
+    def __init__(self, tokenizer: PreTrainedTokenizer, file_dir: str, block_size: int):
+        assert os.path.isdir(file_dir)
+        logger.info(f"Creating features from dataset file folder at {file_dir}")
+        self.examples = []
+        # TODO: randomness could apply a random seed, ex. rng = random.Random(random_seed)
+        # file path looks like ./dataset/wiki_1, ./dataset/wiki_2
+        for file_name in os.listdir(file_dir):
+            file_path = os.path.join(file_dir, file_name)
+            assert os.path.isfile(file_path)
+            article_open = False
+            with open(file_path, encoding="utf-8") as f:
+                original_lines = f.readlines()
+                article_lines = []
+                for line in original_lines:
+                    if "<doc id=" in line:
+                        article_open = True
+                    elif "</doc>" in line:
+                        article_open = False
+                        document = [
+                            tokenizer.convert_tokens_to_ids(tokenizer.tokenize(line))
+                            for line in article_lines[1:]
+                            if (len(line) > 0 and not line.isspace())
+                        ]
+
+                        examples = self.create_examples_from_document(document, block_size, tokenizer)
+                        self.examples.extend(examples)
+                        article_lines = []
+                    else:
+                        if article_open:
+                            article_lines.append(line)
+
+        logger.info("Dataset parse finished.")
+
+    def create_examples_from_document(self, document, block_size, tokenizer, short_seq_prob=0.1):
+        """Creates examples for a single document."""
+
+        # Account for special tokens
+        max_num_tokens = block_size - tokenizer.num_special_tokens_to_add(pair=True)
+
+        # We *usually* want to fill up the entire sequence since we are padding
+        # to `block_size` anyways, so short sequences are generally wasted
+        # computation. However, we *sometimes*
+        # (i.e., short_seq_prob == 0.1 == 10% of the time) want to use shorter
+        # sequences to minimize the mismatch between pre-training and fine-tuning.
+        # The `target_seq_length` is just a rough target however, whereas
+        # `block_size` is a hard limit.
+        target_seq_length = max_num_tokens
+        if random.random() < short_seq_prob:
+            target_seq_length = random.randint(2, max_num_tokens)
+
+        # We DON'T just concatenate all of the tokens from a document into a long
+        # sequence and choose an arbitrary split point because this would make the
+        # next sentence prediction task too easy. Instead, we split the input into
+        # segments "A" and "B" based on the actual "sentences" provided by the user
+        # input.
+        examples = []
+        current_chunk = []  # a buffer stored current working segments
+        current_length = 0
+        i = 0
+        while i < len(document):
+            segment = document[i]  # get a segment
+            if not segment:
+                i += 1
+                continue
+            current_chunk.append(segment)  # add a segment to current chunk
+            current_length += len(segment)  # overall token length
+            # if current length goes to the target length or reaches the end of file, start building token a and b
+            if i == len(document) - 1 or current_length >= target_seq_length:
+                if current_chunk:
+                    # `a_end` is how many segments from `current_chunk` go into the `A` (first) sentence.
+                    a_end = 1
+                    # if current chunk has more than 2 sentences, pick part of it `A` (first) sentence
+                    if len(current_chunk) >= 2:
+                        a_end = random.randint(1, len(current_chunk) - 1)
+                    # token a
+                    tokens_a = []
+                    for j in range(a_end):
+                        tokens_a.extend(current_chunk[j])
+
+                    # token b
+                    tokens_b = []
+                    for j in range(a_end, len(current_chunk)):
+                        tokens_b.extend(current_chunk[j])
+
+                    if len(tokens_a) == 0 or len(tokens_b) == 0:
+                        continue
+
+                    # switch tokens_a and tokens_b randomly
+                    if random.random() < 0.5:
+                        is_next = False
+                        tokens_a, tokens_b = tokens_b, tokens_a
+                    else:
+                        is_next = True
+
+                    def truncate_seq_pair(tokens_a, tokens_b, max_num_tokens):
+                        """Truncates a pair of sequences to a maximum sequence length."""
+                        while True:
+                            total_length = len(tokens_a) + len(tokens_b)
+                            if total_length <= max_num_tokens:
+                                break
+                            trunc_tokens = tokens_a if len(tokens_a) > len(tokens_b) else tokens_b
+                            assert len(trunc_tokens) >= 1
+                            # We want to sometimes truncate from the front and sometimes from the
+                            # back to add more randomness and avoid biases.
+                            if random.random() < 0.5:
+                                del trunc_tokens[0]
+                            else:
+                                trunc_tokens.pop()
+
+                    truncate_seq_pair(tokens_a, tokens_b, max_num_tokens)
+                    assert len(tokens_a) >= 1
+                    assert len(tokens_b) >= 1
+
+                    # add special tokens
+                    input_ids = tokenizer.build_inputs_with_special_tokens(tokens_a, tokens_b)
+                    # add token type ids, 0 for sentence a, 1 for sentence b
+                    token_type_ids = tokenizer.create_token_type_ids_from_sequences(tokens_a, tokens_b)
+
+                    example = {
+                        "input_ids": torch.tensor(input_ids, dtype=torch.long),
+                        "token_type_ids": torch.tensor(token_type_ids, dtype=torch.long),
+                        "sentence_order_label": torch.tensor(0 if is_next else 1, dtype=torch.long),
+                    }
+                    examples.append(example)
+                current_chunk = []  # clear current chunk
+                current_length = 0  # reset current text length
+            i += 1  # go to next line
+        return examples
+
+    def __len__(self):
+        return len(self.examples)
+
+    def __getitem__(self, i) -> Dict[str, torch.tensor]:
+        return self.examples[i]
+
+
 class TextDatasetForNextSentencePrediction(Dataset):
     """
     This will be superseded by a framework-agnostic approach

tests/fixtures/tests_samples/wiki_text/wiki_00

+<doc id="12" url="https://en.wikipedia.org/wiki?curid=12" title="Anarchism">
+Anarchism
+
+Anarchism is a political philosophy and movement that rejects all involuntary, coercive forms of hierarchy. It radically calls for the abolition of the state which it holds to be undesirable, unnecessary, and harmful.
+
+The history of anarchism stretches back to prehistory, when humans lived in anarchistic societies long before the establishment of formal states, realms or empires. With the rise of organised hierarchical bodies, skepticism toward authority also rose, but it was not until the 19th century that a self-conscious political movement emerged. During the latter half of the 19th and the first decades of the 20th century, the anarchist movement flourished in most parts of the world and had a significant role in worker's struggles for emancipation. Various anarchist schools of thought formed during this period.
+
+Anarchists took part in several revolutions, most notably in the Spanish Civil War, where they were crushed along with the alliance to restore the Second Republic by the fascist forces of the Nationalist faction and its foreign allies in Nazi Germany, Fascist Italy, Portuguese Dictatorship and the Catholic Church in 1939, marking the end of the classical era of anarchism. In the last decades of the 20th century and into the 21st century, the anarchist movement has been resurgent once more.
+
+Anarchism employs various tactics in order to meet its ideal ends; these can be broadly separated into revolutionary and evolutionary tactics. There is significant overlap between the two, which are merely descriptive. Revolutionary tactics aim to bring down authority and state, and have taken a violent turn in the past. Evolutionary tactics aim to prefigure what an anarchist society would be like. Anarchist thought, criticism, and praxis has played a part in diverse areas of human society.
+
+The etymological origin of "anarchism" is from the Ancient Greek "anarkhia", meaning "without a ruler", composed of the prefix "an-" (i.e. "without") and the word "arkhos" (i.e. "leader" or "ruler"). The suffix "-ism" denotes the ideological current that favours anarchy. "Anarchism" appears in English from 1642 as "anarchisme" and "anarchy" from 1539. Various factions within the French Revolution labelled their opponents as "anarchists", although few such accused shared many views with later anarchists. Many revolutionaries of the 19th century such as William Godwin (1756–1836) and Wilhelm Weitling (1808–1871) would contribute to the anarchist doctrines of the next generation, but they did not use "anarchist" or "anarchism" in describing themselves or their beliefs.
+
+The first political philosopher to call himself an "anarchist" () was Pierre-Joseph Proudhon (1809–1865), marking the formal birth of anarchism in the mid-19th century. Since the 1890s and beginning in France, "libertarianism" has often been used as a synonym for anarchism and its use as a synonym is still common outside the United States. On the other hand, some use "libertarianism" to refer to individualistic free-market philosophy only, referring to free-market anarchism as "libertarian anarchism".
+
+While opposition to the state is central to anarchist thought, defining anarchism is not an easy task as there is a lot of discussion among scholars and anarchists on the matter and various currents perceive anarchism slightly differently. Hence, it might be true to say that anarchism is a cluster of political philosophies opposing authority and hierarchical organization (including the state, capitalism, nationalism and all associated institutions) in the conduct of all human relations in favour of a society based on voluntary association, on freedom and on decentralisation, but this definition has the same shortcomings as the definition based on etymology (which is simply a negation of a ruler), or based on anti-statism (anarchism is much more than that) or even the anti-authoritarian (which is an "a posteriori" conclusion). Nonetheless, major elements of the definition of anarchism include the following:
+
+During the prehistoric era of mankind, an established authority did not exist. It was after the creation of towns and cities that institutions of authority were established and anarchistic ideas espoused as a reaction. Most notable precursors to anarchism in the ancient world were in China and Greece. In China, philosophical anarchism (i.e. the discussion on the legitimacy of the state) was delineated by Taoist philosophers Zhuang Zhou and Laozi.
+
+Likewise, anarchic attitudes were articulated by tragedians and philosophers in Greece. Aeschylus and Sophocles used the myth of Antigone to illustrate the conflict between rules set by the state and personal autonomy. Socrates questioned Athenian authorities constantly and insisted to the right of individual freedom of consciousness. Cynics dismissed human law ("nomos") and associated authorities while trying to live according to nature ("physis"). Stoics were supportive of a society based on unofficial and friendly relations among its citizens without the presence of a state.
+
+During the Middle Ages, there was no anarchistic activity except some ascetic religious movements in the Muslim world or in Christian Europe. This kind of tradition later gave birth to religious anarchism. In the Sasanian Empire, Mazdak called for an egalitarian society and the abolition of monarchy, only to be soon executed by Emperor Kavad I.
+
+In Basra, religious sects preached against the state. In Europe, various sects developed anti-state and libertarian tendencies. Libertarian ideas further emerged during the Renaissance with the spread of reasoning and humanism through Europe. Novelists fictionalised ideal societies that were based not on coercion but voluntarism. The Enlightenment further pushed towards anarchism with the optimism for social progress.
+
+During the French Revolution, partisan groups such as the Enragés and the saw a turning point in the fermentation of anti-state and federalist sentiments. The first anarchist currents developed throughout the 18th century—William Godwin espoused philosophical anarchism in England, morally delegitimizing the state, Max Stirner's thinking paved the way to individualism, and Pierre-Joseph Proudhon's theory of mutualism found fertile soil in France. This era of classical anarchism lasted until the end of the Spanish Civil War of 1936 and is considered the golden age of anarchism.
+Drawing from mutualism, Mikhail Bakunin founded collectivist anarchism and entered the International Workingmen's Association, a class worker union later known as the First International that formed in 1864 to unite diverse revolutionary currents. The International became a significant political force, with Karl Marx being a leading figure and a member of its General Council. Bakunin's faction (the Jura Federation) and Proudhon's followers (the mutualists) opposed Marxist state socialism, advocating political abstentionism and small property holdings. After bitter disputes, the Bakuninists were expelled from the International by the Marxists at the 1872 Hague Congress. Bakunin famously predicted that if revolutionaries gained power by Marx's terms, they would end up the new tyrants of workers. After being expelled, anarchists formed the St. Imier International. Under the influence of Peter Kropotkin, a Russian philosopher and scientist, anarcho-communism overlapped with collectivism. Anarcho-communists, who drew inspiration from the 1871 Paris Commune, advocated for free federation and for the distribution of goods according to one's needs.
+
+At the turn of the century, anarchism had spread all over the world. In China, small groups of students imported the humanistic pro-science version of anarcho-communism. Tokyo was a hotspot for rebellious youth from countries of the far east, travelling to the Japanese capital to study. In Latin America, Argentina was a stronghold for anarcho-syndicalism, where it became the most prominent left-wing ideology. During this time, a minority of anarchists adopted tactics of revolutionary political violence. This strategy became known as propaganda of the deed. The dismemberment of the French socialist movement into many groups, and the execution and exile of many Communards to penal colonies following the suppression of the Paris Commune, favoured individualist political expression and acts. Even though many anarchists distanced themselves from these terrorist acts, infamy came upon the movement. Illegalism was another strategy which some anarchists adopted during this period.
+Anarchists enthusiastically participated in the Russian Revolution—despite concerns—in opposition to the Whites. However, they met harsh suppression after the Bolshevik government was stabilized. Several anarchists from Petrograd and Moscow fled to Ukraine, notably leading to the Kronstadt rebellion and Nestor Makhno's struggle in the Free Territory. With the anarchists being crushed in Russia, two new antithetical currents emerged, namely platformism and synthesis anarchism. The former sought to create a coherent group that would push for revolution while the latter were against anything that would resemble a political party. Seeing the victories of the Bolsheviks in the October Revolution and the resulting Russian Civil War, many workers and activists turned to communist parties, which grew at the expense of anarchism and other socialist movements. In France and the United States, members of major syndicalist movements, the General Confederation of Labour and Industrial Workers of the World, left their organisations and joined the Communist International.
+
+In the Spanish Civil War, anarchists and syndicalists (CNT and FAI) once again allied themselves with various currents of leftists. A long tradition of Spanish anarchism led to anarchists playing a pivotal role in the war. In response to the army rebellion, an anarchist-inspired movement of peasants and workers, supported by armed militias, took control of Barcelona and of large areas of rural Spain, where they collectivised the land. The Soviet Union provided some limited assistance at the beginning of the war, but the result was a bitter fight among communists and anarchists at a series of events named May Days as Joseph Stalin tried to seize control of the Republicans.
+
+At the end of World War II, the anarchist movement was severely weakened. However, the 1960s witnessed a revival of anarchism likely caused by a perceived failure of Marxism–Leninism and tensions built by the Cold War. During this time, anarchism took root in other movements critical towards both the state and capitalism, such as the anti-nuclear, environmental and pacifist movements, the New Left, and the counterculture of the 1960s. Anarchism became associated with punk subculture, as exemplified by bands such as Crass and the Sex Pistols, and the established feminist tendencies of anarcha-feminism returned with vigour during the second wave of feminism.
+
+Around the turn of the 21st century, anarchism grew in popularity and influence within anti-war, anti-capitalist, and anti-globalisation movements. Anarchists became known for their involvement in protests against the World Trade Organization, the Group of Eight and the World Economic Forum. During the protests, "ad hoc" leaderless anonymous cadres known as black blocs engaged in rioting, property destruction, and violent confrontations with the police. Other organisational tactics pioneered in this time include security culture, affinity groups, and the use of decentralised technologies such as the internet. A significant event of this period was the confrontations at the WTO conference in Seattle in 1999. Anarchist ideas have been influential in the development of the Zapatistas in Mexico and the Democratic Federation of Northern Syria, more commonly known as Rojava, a "de facto" autonomous region in northern Syria.
+
+Anarchist schools of thought have been generally grouped into two main historical traditions, social anarchism and individualist anarchism, owing to their different origins, values and evolution. The individualist current emphasises negative liberty in opposing restraints upon the free individual, while the social current emphasises positive liberty in aiming to achieve the free potential of society through equality and social ownership. In a chronological sense, anarchism can be segmented by the classical currents of the late 19th century, and the post-classical currents (such as anarcha-feminism, green anarchism and post-anarchism) developed thereafter.
+
+Beyond the specific factions of anarchist movements which constitute political anarchism lies philosophical anarchism, which holds that the state lacks moral legitimacy, without necessarily accepting the imperative of revolution to eliminate it. A component especially of individualist anarchism, philosophical anarchism may tolerate the existence of a minimal state, but argues that citizens have no moral obligation to obey government when it conflicts with individual autonomy. Anarchism pays significant attention to moral arguments since ethics have a central role in anarchist philosophy.
+
+One reaction against sectarianism within the anarchist milieu was anarchism without adjectives, a call for toleration and unity among anarchists first adopted by Fernando Tarrida del Mármol in 1889 in response to the bitter debates of anarchist theory at the time. Despite separation, the various anarchist schools of thought are not seen as distinct entities, but as tendencies that intermingle.
+
+Anarchism is usually placed on the far-left of the political spectrum. Much of its economics and legal philosophy reflect anti-authoritarian, anti-statist, and libertarian interpretations of the radical left-wing and socialist politics of collectivism, communism, individualism, mutualism, and syndicalism, among other libertarian socialist economic theories. As anarchism does not offer a fixed body of doctrine from a single particular worldview, many anarchist types and traditions exist, and varieties of anarchy diverge widely.
+
+Inceptive currents among classical anarchist currents were mutualism and individualism. They were followed by the major currents of social anarchism (collectivist, communist, and syndicalist). They differ on organizational and economic aspects of their ideal society.
+
+Mutualism is an 18th-century economic theory that was developed into anarchist theory by Pierre-Joseph Proudhon. Its aims include reciprocity, free association, voluntary contract, federation, and credit and currency reform that would be regulated by a bank of the people. Mutualism has been retrospectively characterised as ideologically situated between individualist and collectivist forms of anarchism. Proudhon first characterised his goal as a "third form of society, the synthesis of communism and property".
+
+Collectivist anarchism, also known as anarchist collectivism or anarcho-collectivism, is a revolutionary socialist form of anarchism commonly associated with Mikhail Bakunin. Collectivist anarchists advocate collective ownership of the means of production, theorised to be achieved through violent revolution, and that workers be paid according to time worked, rather than goods being distributed according to need as in communism. Collectivist anarchism arose alongside Marxism, but rejected the dictatorship of the proletariat despite the stated Marxist goal of a collectivist stateless society. Anarcho-communism, also known as anarchist-communism, communist anarchism, and libertarian communism, is a theory of anarchism that advocates a communist society with common ownership of the means of production, direct democracy, and a horizontal network of voluntary associations and workers' councils with production and consumption based on the guiding principle: "From each according to his ability, to each according to his need". Anarcho-communism developed from radical socialist currents after the French Revolution, but it was first formulated as such in the Italian section of the First International. It was later expanded upon in the theoretical work of Peter Kropotkin.
+
+Anarcho-syndicalism, also referred to as revolutionary syndicalism, is a branch of anarchism that views labour syndicates as a potential force for revolutionary social change, replacing capitalism and the state with a new society democratically self-managed by workers. The basic principles of anarcho-syndicalism are workers' solidarity, direct action, and workers' self-management.
+
+Individualist anarchism refers to several traditions of thought within the anarchist movement that emphasise the individual and their will over any kinds of external determinants. Early influences on individualist forms of anarchism include William Godwin, Max Stirner and Henry David Thoreau. Through many countries, individualist anarchism attracted a small yet diverse following of Bohemian artists and intellectuals as well as young anarchist outlaws in what became known as illegalism and individual reclamation.
+
+Anarchist principles undergird contemporary radical social movements of the left. Interest in the anarchist movement developed alongside momentum in the anti-globalization movement, whose leading activist networks were anarchist in orientation. As the movement shaped 21st century radicalism, wider embrace of anarchist principles signaled a revival of interest. Contemporary news coverage which emphasizes black bloc demonstrations has reinforced anarchism's historical association with chaos and violence, although its publicity has also led more scholars to engage with the anarchist movement. Anarchism has continued to generate many philosophies and movements—at times eclectic, drawing upon various sources, and syncretic, combining disparate concepts to create new philosophical approaches. The anti-capitalist tradition of classical anarchism has remained prominent within contemporary currents.
+
+Various anarchist groups, tendencies, and schools of thought exist today, making it difficult to describe contemporary anarchist movement. While theorists and activists have established "relatively stable constellations of anarchist principles", there is no consensus on which principles are core. As a result, commentators describe multiple "anarchisms" (rather than a singular "anarchism") in which common principles are shared between schools of anarchism while each group prioritizes those principles differently. For example, gender equality can be a common principle but ranks as a higher priority to anarcha-feminists than anarchist communists. Anarchists are generally committed against coercive authority in all forms, namely "all centralized and hierarchical forms of government (e.g., monarchy, representative democracy, state socialism, etc.), economic class systems (e.g., capitalism, Bolshevism, feudalism, slavery, etc.), autocratic religions (e.g., fundamentalist Islam, Roman Catholicism, etc.), patriarchy, heterosexism, white supremacy, and imperialism". However, anarchist schools disagree on the methods by which these forms should be opposed.
+
+Anarchists' tactics take various forms but in general serve two major goals—first, to oppose the Establishment; and second, to promote anarchist ethics and reflect an anarchist vision of society, illustrating the unity of means and ends. A broad categorization can be made between aims to destroy oppressive states and institutions by revolutionary means, and aims to change society through evolutionary means. Evolutionary tactics reject violence and take a gradual approach to anarchist aims, though there is significant overlap between the two.
+
+Anarchist tactics have shifted during the course of the last century. Anarchists during the early 20th century focused more on strikes and militancy, while contemporary anarchists use a broader array of approaches.
+
+During the classical era, anarchists had a militant tendency. Not only did they confront state armed forces (as in Spain and Ukraine) but some of them also employed terrorism as propaganda of the deed. Assassination attempts were carried out against heads of state, some of which were successful. Anarchists also took part in revolutions. Anarchist perspectives towards violence have always been perplexing and controversial. On one hand, anarcho-pacifists point out the unity of means and ends. On the other hand, other anarchist groups advocate direct action, a tactic which can include acts of sabotage or even acts of terrorism. This attitude was quite prominent a century ago; seeing the state as a tyrant, some anarchists believed that they had every right to oppose its oppression by any means possible. Emma Goldman and Errico Malatesta, who were proponents of limited use of violence, argued that violence is merely a reaction to state violence as a necessary evil.
+
+Anarchists took an active role in strikes, although they tended to be antipathetic to formal syndicalism, seeing it as reformist. They saw it as a part of the movement which sought to overthrow the state and capitalism. Anarchists also reinforced their propaganda within the arts, some of whom practiced nudism. They also built communities which were based on friendship. They were also involved in the press.
+
+In the current era, Italian anarchist Alfredo Bonanno, a proponent of insurrectionary anarchism, has reinstated the debate on violence by rejecting the nonviolence tactic adopted since the late 19th century by Kropotkin and other prominent anarchists afterwards. Both Bonanno and the French group The Invisible Committee advocate for small, informal affiliation groups, where each member is responsible for their own actions but works together to bring down oppression utilizing sabotage and other violent means against state, capitalism and other enemies. Members of The Invisible Committee were arrested in 2008 on various charges, terrorism included.
+
+Overall, today's anarchists are much less violent and militant than their ideological ancestors. They mostly engage in confronting the police during demonstrations and riots, especially in countries like Canada, Mexico or Greece. Μilitant black bloc protest groups are known for clashing with the police. However, anarchists not only clash with state operators; they also engage in the struggle against fascists and racists, taking anti-fascist action and mobilizing to prevent hate rallies from happening.
+
+Anarchists commonly employ direct action. This can take the form of disrupting and protesting against unjust hierarchy, or the form of self-managing their lives through the creation of counter-institutions such as communes and non-hierarchical collectives. Often, decision-making is handled in an anti-authoritarian way, with everyone having equal say in each decision, an approach known as horizontalism. Contemporary-era anarchists have been engaging with various grassroots movements that are not explicitly anarchist but are more or less based on horizontalism, respecting personal autonomy, and participating in mass activism such as strikes and demonstrations. The newly coined term "small-a anarchism", in contrast with the "big-A anarchism" of the classical era, signals their tendency not to base their thoughts and actions on classical-era anarchism or to refer to Kropotkin or Proudhon to justify their opinions. They would rather base their thought and praxis on their own experience, which they will later theorize.
+
+The decision-making process of small affinity anarchist groups play a significant tactical role. Anarchists have employed various methods in order to build a rough consensus among members of their group, without the need of a leader or a leading group. One way is for an individual from the group to play the role of facilitator to help achieve a consensus without taking part in the discussion themselves or promoting a specific point. Minorities usually accept rough consensus, except when they feel the proposal contradicts anarchist goals, values, or ethics. Anarchists usually form small groups (5–20 individuals) to enhance autonomy and friendships among their members. These kind of groups more often than not interconnect with each other, forming larger networks. Anarchists still support and participate in strikes, especially wildcat strikes; these are leaderless strikes not organised centrally by a syndicate.
+
+Anarchists have gone online to spread their message. As in the past, newspapers and journals are used; however, because of distributional and other difficulties, anarchists have found it easier to create websites, hosting electronic libraries and other portals. Anarchists were also involved in developing various software that are available for free. The way these hacktivists work to develop and distribute resembles the anarchist ideals, especially when it comes to preserving user's privacy from state surveillance.
+
+Anarchists organize themselves to squat and reclaim public spaces. During important events such as protests and when spaces are being occupied, they are often called Temporary Autonomous Zones (TAZ), spaces where surrealism, poetry and art are blended to display the anarchist ideal. As seen by anarchists, squatting is a way to regain urban space from the capitalist market, serving pragmatical needs, and is also seen an exemplary direct action. Acquiring space enables anarchists to experiment with their ideas and build social bonds. Adding up these tactics, and having in mind that not all anarchists share the same attitudes towards them, along with various forms of protesting at highly symbolic events, make up a carnivalesque atmosphere that is part of contemporary anarchist vividity.
+
+As anarchism is a philosophy that embodies many diverse attitudes, tendencies, and schools of thought, and disagreement over questions of values, ideology, and tactics is common, its diversity has led to widely different uses of identical terms among different anarchist traditions, which has created a number of definitional concerns in anarchist theory. For instance, the compatibility of capitalism, nationalism and religion with anarchism is widely disputed. Similarly, anarchism enjoys complex relationships with ideologies such as Marxism, communism, collectivism and trade unionism. Anarchists may be motivated by humanism, divine authority, enlightened self-interest, veganism, or any number of alternative ethical doctrines. Phenomena such as civilisation, technology (e.g. within anarcho-primitivism) and the democratic process may be sharply criticised within some anarchist tendencies and simultaneously lauded in others.
+
+Gender and sexuality carry along them dynamics of hierarchy; anarchism is obliged to address, analyse and oppose the suppression of one's autonomy because of the dynamics that gender roles traditionally impose.
+
+A historical current that arose and flourished during 1890 and 1920 within anarchism was free love; in contemporary anarchism, this current survives as a tendency to support polyamory and queer anarchism. Free love advocates were against marriage, which they saw as a way of men imposing authority over women, largely because marriage law greatly favoured the power of men. The notion of free love, though, was much broader; it included critique of the established order that limited women's sexual freedom and pleasure. Such free love movements contributed to the establishment of communal houses, where large groups of travelers, anarchists, and other activists slept in beds together. Free love had roots both in Europe and the United States. Some anarchists, however, struggled with the jealousy that arose from free love. Anarchist feminists were advocates of free love, against marriage, were pro-choice (utilizing a contemporary term) and had a likewise agenda. Anarchist and non-anarchist feminists differed on suffrage, but were nonetheless supportive of one another.
+
+During the second half of the 20th century, anarchism intermingled with the second wave of feminism, radicalizing some currents of the feminist movement (and being influenced as well). By the latest decades of the 20th century, anarchists and feminists were advocating for the rights and autonomy of women, gays, queers and other marginalized groups, with some feminist thinkers suggesting a fusion of the two currents. With the third wave of feminism, sexual identity and compulsory heterosexuality became a subject of study for anarchists, which yielded a post-structuralist critique of sexual normality. However, some anarchists distanced themselves from this line of thinking, suggesting that it leaned towards individualism and was, therefore, dropping the cause of social liberation.
+
+The interest of anarchists in education stretches back to the first emergence of classical anarchism. Anarchists consider 'proper' education, which sets the foundations of the future autonomy of the individual and the society, to be an act of mutual aid. Anarchist writers such as Willian Godwin and Max Stirner attacked both state education and private education as another means by which the ruling class replicate their privileges.
+
+In 1901, Catalan anarchist and free thinker Francisco Ferrer established the Escuela Moderna in Barcelona as an opposition to the established education system, which was dictated largely by the Catholic Church. Ferrer's approach was secular, rejecting both state and church involvement in the educational process, and gave pupils large amounts of autonomy in planning their work and attendance. Ferrer aimed to educate the working class and explicitly sought to foster class consciousness among students. The school closed after constant harassment by the state and Ferrer was later arrested. His ideas, however, formed the inspiration for a series of modern schools around the world. Christian anarchist Leo Tolstoy also established a similar school, with its founding principle, according to Tolstoy, being that "for education to be effective it had to be free". In a similar token, A. S. Neill founding what became Summerhill School in 1921, also declaring being free from coercion. 
+
+Anarchist education is based largely on the idea that a child's right to develop freely, without manipulation, ought to be respected, and that rationality will lead children to morally good conclusions. However, there has been little consensus among anarchist figures as to what constitutes manipulation; Ferrer, for example, believed that moral indoctrination was necessary and explicitly taught pupils that equality, liberty, and social justice were not possible under capitalism (along with other critiques of nationalism and government). 
+
+Late 20th century and contemporary anarchist writers (such as Colin Ward, Herbert Read and Paul Goodman) intensified and expanded the anarchist critique of state education, largely focusing on the need for a system that focuses on children's creativity rather than on their ability to attain a career or participate in consumer society. Contemporary anarchists, such as Colin Ward, have further argued that state education serves to perpetuate socio-economic inequality.
+
+While few anarchist education institutions have survived to the modern day, major tenets of anarchist schools, such as respect for child autonomy and relying on reasoning rather than indoctrination as a teaching method, have spread among mainstream educational institutions.
+
+Objection to the state and its institutions is a "sine qua non" of anarchism. Anarchists consider the state as a tool of domination and believe it to be illegitimate regardless of its political tendencies. Instead of people being able to control the aspects of their life, major decisions are taken by a small elite. Authority ultimately rests solely on power, regardless of whether that power is open or transparent, as it still has the ability to coerce people. Another anarchist argument against states is that the people constituting a government, even the most altruistic among officials, will unavoidably seek to gain more power, leading to corruption. Anarchists consider the idea that the state is the collective will of the people to be an unachievable fiction, due to the fact that the ruling class is distinct from the rest of society.
+
+The connection between anarchism and art was quite profound during the classical era of anarchism, especially among artistic currents that were developing during that era, such as futurists, surrealists, and others, while in literature anarchism was mostly associated with the New Apocalyptics and the Neo-romanticism movement. In music, anarchism has been associated with music scenes such as Punk. Anarchists such as Leo Tolstoy and Herbert Read argued that the border between the artist and the non-artist, what separates art from a daily act, is a construct produced by the alienation caused by capitalism, and it prevents humans from living a joyful life. 
+
+Other anarchists advocated for or used art as a means to achieve anarchist ends. In his book Breaking the Spell: A History of Anarchist Filmmakers, Videotape Guerrillas, and Digital Ninjas Chris Robé claims that "anarchist-inflected practices have increasingly structured movement-based video activism." 
+
+Three overlapping properties made art useful to anarchists: It could depict a critique of existing society and hierarchies; it could serve as a prefigurative tool to reflect the anarchist ideal society, and also it could turn into a means of direct action, in protests for example. As it appeals to both emotion and reason, art could appeal to the "whole human" and have a powerful effect.
+
+Philosophy lecturer Andrew G. Fiala has listed five main arguments against anarchism. Firstly, he notes that anarchism is related to violence and destruction, not only in the pragmatic world (i.e. at protests) but in the world of ethics as well. The second argument is that it is impossible for a society to function without a state or something like a state, acting to protect citizens from criminality. Fiala takes "Leviathan" from Thomas Hobbes and the night-watchman state from philosopher Robert Nozick as examples. Thirdly, anarchism is evaluated as unfeasible or utopian since the state can not be defeated practically; this line of arguments most often calls for political action within the system to reform it. The fourth argument is that anarchism is self-contradictory since while it advocates for no-one to "archiei", if accepted by the many, then anarchism will turn into the ruling political theory. In this line of criticism also comes the self contradiction that anarchist calls for collective action while anarchism endorses the autonomy of the individual and hence no collective action can be taken. Lastly, Fiala mentions a critique towards philosophical anarchism, of being ineffective (all talk and thoughts) and in the meantime capitalism and bourgeois class remains strong.
+
+Philosophical anarchism has met the criticism of members of academia, following the release of pro-anarchist books such as A. John Simmons' "Moral Principles and Political Obligations" (1979). Law professor William A. Edmundson authored an essay arguing against three major philosophical anarchist principles, which he finds fallacious; Edmundson claims that while the individual does not owe a normal state a duty of obedience, this does not imply that anarchism is the inevitable conclusion, and the state is still morally legitimate.
+
+
+
+
+
+
+
+</doc>
+<doc id="25" url="https://en.wikipedia.org/wiki?curid=25" title="Autism">
+Autism
+
+Autism is a developmental disorder characterized by difficulties with social interaction and communication, and by restricted and repetitive behavior. Parents often notice signs during the first three years of their child's life. These signs often develop gradually, though some children with autism experience worsening in their communication and social skills after reaching developmental milestones at a normal pace.
+Autism is associated with a combination of genetic and environmental factors. Risk factors during pregnancy include certain infections, such as rubella, toxins including valproic acid, alcohol, cocaine, pesticides, lead, and air pollution, fetal growth restriction, and autoimmune diseases. Controversies surround other proposed environmental causes; for example, the vaccine hypothesis, which has been disproven. Autism affects information processing in the brain and how nerve cells and their synapses connect and organize; how this occurs is not well understood. The Diagnostic and Statistical Manual of Mental Disorders (DSM-5), combines autism and less severe forms of the condition, including Asperger syndrome and pervasive developmental disorder not otherwise specified (PDD-NOS) into the diagnosis of autism spectrum disorder (ASD).
+Early behavioral interventions or speech therapy can help children with autism gain self-care, social, and communication skills. Although there is no known cure, there have been cases of children who recovered. Some autistic adults are unable to live independently. An autistic culture has developed, with some individuals seeking a cure and others believing autism should be accepted as a difference to be accommodated instead of cured.
+Globally, autism is estimated to affect 24.8 million people . In the 2000s, the number of people affected was estimated at 1–2 per 1,000 people worldwide. In the developed countries, about 1.5% of children are diagnosed with ASD , from 0.7% in 2000 in the United States. It occurs four-to-five times more often in males than females. The number of people diagnosed has increased dramatically since the 1960s, which may be partly due to changes in diagnostic practice. The question of whether actual rates have increased is unresolved.
+Autism is a highly variable, neurodevelopmental disorder whose symptoms first appears during infancy or childhood, and generally follows a steady course without remission. People with autism may be severely impaired in some respects but average, or even superior, in others. Overt symptoms gradually begin after the age of six months, become established by age two or three years and tend to continue through adulthood, although often in more muted form. It is distinguished by a characteristic triad of symptoms: impairments in social interaction, impairments in communication, and repetitive behavior. Other aspects, such as atypical eating, are also common but are not essential for diagnosis. Individual symptoms of autism occur in the general population and appear not to associate highly, without a sharp line separating pathologically severe from common traits.
+
+Social deficits distinguish autism and the related autism spectrum disorders (ASD; see Classification) from other developmental disorders. People with autism have social impairments and often lack the intuition about others that many people take for granted. Noted autistic Temple Grandin described her inability to understand the social communication of neurotypicals, or people with typical neural development, as leaving her feeling "like an anthropologist on Mars".
+
+Unusual social development becomes apparent early in childhood. Autistic infants show less attention to social stimuli, smile and look at others less often, and respond less to their own name. Autistic toddlers differ more strikingly from social norms; for example, they have less eye contact and turn-taking, and do not have the ability to use simple movements to express themselves, such as pointing at things. Three- to five-year-old children with autism are less likely to exhibit social understanding, approach others spontaneously, imitate and respond to emotions, communicate nonverbally, and take turns with others. However, they do form attachments to their primary caregivers. Most children with autism display moderately less attachment security than neurotypical children, although this difference disappears in children with higher mental development or less pronounced autistic traits. Older children and adults with ASD perform worse on tests of face and emotion recognition although this may be partly due to a lower ability to define a person's own emotions.
+
+Children with high-functioning autism have more intense and frequent loneliness compared to non-autistic peers, despite the common belief that children with autism prefer to be alone. Making and maintaining friendships often proves to be difficult for those with autism. For them, the quality of friendships, not the number of friends, predicts how lonely they feel. Functional friendships, such as those resulting in invitations to parties, may affect the quality of life more deeply.
+There are many anecdotal reports, but few systematic studies, of aggression and violence in individuals with ASD. The limited data suggest that, in children with intellectual disability, autism is associated with aggression, destruction of property, and meltdowns.
+
+About a third to a half of individuals with autism do not develop enough natural speech to meet their daily communication needs. Differences in communication may be present from the first year of life, and may include delayed onset of babbling, unusual gestures, diminished responsiveness, and vocal patterns that are not synchronized with the caregiver. In the second and third years, children with autism have less frequent and less diverse babbling, consonants, words, and word combinations; their gestures are less often integrated with words. Children with autism are less likely to make requests or share experiences, and are more likely to simply repeat others' words (echolalia) or reverse pronouns. Joint attention seems to be necessary for functional speech, and deficits in joint attention seem to distinguish infants with ASD. For example, they may look at a pointing hand instead of the pointed-at object, and they consistently fail to point at objects in order to comment on or share an experience. Children with autism may have difficulty with imaginative play and with developing symbols into language.
+
+In a pair of studies, high-functioning children with autism aged 8–15 performed equally well as, and as adults better than, individually matched controls at basic language tasks involving vocabulary and spelling. Both autistic groups performed worse than controls at complex language tasks such as figurative language, comprehension and inference. As people are often sized up initially from their basic language skills, these studies suggest that people speaking to autistic individuals are more likely to overestimate what their audience comprehends.
+
+Autistic individuals can display many forms of repetitive or restricted behavior, which the Repetitive Behavior Scale-Revised (RBS-R) categorizes as follows.
+
+
+No single repetitive or self-injurious behavior seems to be specific to autism, but autism appears to have an elevated pattern of occurrence and severity of these behaviors.
+
+Autistic individuals may have symptoms that are independent of the diagnosis, but that can affect the individual or the family.
+An estimated 0.5% to 10% of individuals with ASD show unusual abilities, ranging from splinter skills such as the memorization of trivia to the extraordinarily rare talents of prodigious autistic savants. Many individuals with ASD show superior skills in perception and attention, relative to the general population. Sensory abnormalities are found in over 90% of those with autism, and are considered core features by some, although there is no good evidence that sensory symptoms differentiate autism from other developmental disorders. Differences are greater for under-responsivity (for example, walking into things) than for over-responsivity (for example, distress from loud noises) or for sensation seeking (for example, rhythmic movements). An estimated 60–80% of autistic people have motor signs that include poor muscle tone, poor motor planning, and toe walking; deficits in motor coordination are pervasive across ASD and are greater in autism proper. Unusual eating behavior occurs in about three-quarters of children with ASD, to the extent that it was formerly a diagnostic indicator. Selectivity is the most common problem, although eating rituals and food refusal also occur.
+
+There is tentative evidence that autism occurs more frequently in people with gender dysphoria.
+
+Gastrointestinal problems are one of the most commonly associated medical disorders in people with autism. These are linked to greater social impairment, irritability, behavior and sleep problems, language impairments and mood changes.
+
+Parents of children with ASD have higher levels of stress. Siblings of children with ASD report greater admiration of and less conflict with the affected sibling than siblings of unaffected children and were similar to siblings of children with Down syndrome in these aspects of the sibling relationship. However, they reported lower levels of closeness and intimacy than siblings of children with Down syndrome; siblings of individuals with ASD have greater risk of negative well-being and poorer sibling relationships as adults.
+
+It has long been presumed that there is a common cause at the genetic, cognitive, and neural levels for autism's characteristic triad of symptoms. However, there is increasing suspicion that autism is instead a complex disorder whose core aspects have distinct causes that often co-occur.
+Autism has a strong genetic basis, although the genetics of autism are complex and it is unclear whether ASD is explained more by rare mutations with major effects, or by rare multigene interactions of common genetic variants. Complexity arises due to interactions among multiple genes, the environment, and epigenetic factors which do not change DNA sequencing but are heritable and influence gene expression. Many genes have been associated with autism through sequencing the genomes of affected individuals and their parents. Studies of twins suggest that heritability is 0.7 for autism and as high as 0.9 for ASD, and siblings of those with autism are about 25 times more likely to be autistic than the general population. However, most of the mutations that increase autism risk have not been identified. Typically, autism cannot be traced to a Mendelian (single-gene) mutation or to a single chromosome abnormality, and none of the genetic syndromes associated with ASDs have been shown to selectively cause ASD. Numerous candidate genes have been located, with only small effects attributable to any particular gene. Most loci individually explain less than 1% of cases of autism. The large number of autistic individuals with unaffected family members may result from spontaneous structural variation—such as deletions, duplications or inversions in genetic material during meiosis. Hence, a substantial fraction of autism cases may be traceable to genetic causes that are highly heritable but not inherited: that is, the mutation that causes the autism is not present in the parental genome. Autism may be underdiagnosed in women and girls due to an assumption that it is primarily a male condition, but genetic phenomena such as imprinting and X linkage have the ability to raise the frequency and severity of conditions in males, and theories have been put forward for a genetic reason why males are diagnosed more often, such as the imprinted brain theory and the extreme male brain theory.
+
+Maternal nutrition and inflammation during preconception and pregnancy influences fetal neurodevelopment. Intrauterine growth restriction is associated with ASD, in both term and preterm infants. Maternal inflammatory and autoimmune diseases may damage fetal tissues, aggravating a genetic problem or damaging the nervous system.
+
+Exposure to air pollution during pregnancy, especially heavy metals and particulates, may increase the risk of autism. Environmental factors that have been claimed without evidence to contribute to or exacerbate autism include certain foods, infectious diseases, solvents, PCBs, phthalates and phenols used in plastic products, pesticides, brominated flame retardants, alcohol, smoking, illicit drugs, vaccines, and prenatal stress. Some, such as the MMR vaccine, have been completely disproven.
+
+Parents may first become aware of autistic symptoms in their child around the time of a routine vaccination. This has led to unsupported theories blaming vaccine "overload", a vaccine preservative, or the MMR vaccine for causing autism. The latter theory was supported by a litigation-funded study that has since been shown to have been "an elaborate fraud". Although these theories lack convincing scientific evidence and are biologically implausible, parental concern about a potential vaccine link with autism has led to lower rates of childhood immunizations, outbreaks of previously controlled childhood diseases in some countries, and the preventable deaths of several children.
+
+Autism's symptoms result from maturation-related changes in various systems of the brain. How autism occurs is not well understood. Its mechanism can be divided into two areas: the pathophysiology of brain structures and processes associated with autism, and the neuropsychological linkages between brain structures and behaviors. The behaviors appear to have multiple pathophysiologies.
+
+There is evidence that gut–brain axis abnormalities may be involved. A 2015 review proposed that immune dysregulation, gastrointestinal inflammation, malfunction of the autonomic nervous system, gut flora alterations, and food metabolites may cause brain neuroinflammation and dysfunction. A 2016 review concludes that enteric nervous system abnormalities might play a role in neurological disorders such as autism. Neural connections and the immune system are a pathway that may allow diseases originated in the intestine to spread to the brain.
+
+Several lines of evidence point to synaptic dysfunction as a cause of autism. Some rare mutations may lead to autism by disrupting some synaptic pathways, such as those involved with cell adhesion. Gene replacement studies in mice suggest that autistic symptoms are closely related to later developmental steps that depend on activity in synapses and on activity-dependent changes. All known teratogens (agents that cause birth defects) related to the risk of autism appear to act during the first eight weeks from conception, and though this does not exclude the possibility that autism can be initiated or affected later, there is strong evidence that autism arises very early in development.
+
+Diagnosis is based on behavior, not cause or mechanism. Under the DSM-5, autism is characterized by persistent deficits in social communication and interaction across multiple contexts, as well as restricted, repetitive patterns of behavior, interests, or activities. These deficits are present in early childhood, typically before age three, and lead to clinically significant functional impairment. Sample symptoms include lack of social or emotional reciprocity, stereotyped and repetitive use of language or idiosyncratic language, and persistent preoccupation with unusual objects. The disturbance must not be better accounted for by Rett syndrome, intellectual disability or global developmental delay. ICD-10 uses essentially the same definition.
+
+Several diagnostic instruments are available. Two are commonly used in autism research: the Autism Diagnostic Interview-Revised (ADI-R) is a semistructured parent interview, and the Autism Diagnostic Observation Schedule (ADOS) uses observation and interaction with the child. The Childhood Autism Rating Scale (CARS) is used widely in clinical environments to assess severity of autism based on observation of children. The Diagnostic interview for social and communication disorders (DISCO) may also be used.
+
+A pediatrician commonly performs a preliminary investigation by taking developmental history and physically examining the child. If warranted, diagnosis and evaluations are conducted with help from ASD specialists, observing and assessing cognitive, communication, family, and other factors using standardized tools, and taking into account any associated medical conditions. A pediatric neuropsychologist is often asked to assess behavior and cognitive skills, both to aid diagnosis and to help recommend educational interventions. A differential diagnosis for ASD at this stage might also consider intellectual disability, hearing impairment, and a specific language impairment such as Landau–Kleffner syndrome. The presence of autism can make it harder to diagnose coexisting psychiatric disorders such as depression.
+
+Clinical genetics evaluations are often done once ASD is diagnosed, particularly when other symptoms already suggest a genetic cause. Although genetic technology allows clinical geneticists to link an estimated 40% of cases to genetic causes, consensus guidelines in the US and UK are limited to high-resolution chromosome and fragile X testing. A genotype-first model of diagnosis has been proposed, which would routinely assess the genome's copy number variations. As new genetic tests are developed several ethical, legal, and social issues will emerge. Commercial availability of tests may precede adequate understanding of how to use test results, given the complexity of autism's genetics. Metabolic and neuroimaging tests are sometimes helpful, but are not routine.
+
+ASD can sometimes be diagnosed by age 14 months, although diagnosis becomes increasingly stable over the first three years of life: for example, a one-year-old who meets diagnostic criteria for ASD is less likely than a three-year-old to continue to do so a few years later. In the UK the National Autism Plan for Children recommends at most 30 weeks from first concern to completed diagnosis and assessment, though few cases are handled that quickly in practice. Although the symptoms of autism and ASD begin early in childhood, they are sometimes missed; years later, adults may seek diagnoses to help them or their friends and family understand themselves, to help their employers make adjustments, or in some locations to claim disability living allowances or other benefits. Girls are often diagnosed later than boys.
+
+Underdiagnosis and overdiagnosis are problems in marginal cases, and much of the recent increase in the number of reported ASD cases is likely due to changes in diagnostic practices. The increasing popularity of drug treatment options and the expansion of benefits has given providers incentives to diagnose ASD, resulting in some overdiagnosis of children with uncertain symptoms. Conversely, the cost of screening and diagnosis and the challenge of obtaining payment can inhibit or delay diagnosis. It is particularly hard to diagnose autism among the visually impaired, partly because some of its diagnostic criteria depend on vision, and partly because autistic symptoms overlap with those of common blindness syndromes or blindisms.
+
+Autism is one of the five pervasive developmental disorders (PDD), which are characterized by widespread abnormalities of social interactions and communication, and severely restricted interests and highly repetitive behavior. These symptoms do not imply sickness, fragility, or emotional disturbance.
+
+Of the five PDD forms, Asperger syndrome is closest to autism in signs and likely causes; Rett syndrome and childhood disintegrative disorder share several signs with autism, but may have unrelated causes; PDD not otherwise specified (PDD-NOS; also called "atypical autism") is diagnosed when the criteria are not met for a more specific disorder. Unlike with autism, people with Asperger syndrome have no substantial delay in language development. The terminology of autism can be bewildering, with autism, Asperger syndrome and PDD-NOS often called the "autism spectrum disorders" (ASD) or sometimes the "autistic disorders", whereas autism itself is often called "autistic disorder", "childhood autism", or "infantile autism". In this article, "autism" refers to the classic autistic disorder; in clinical practice, though, "autism", "ASD", and "PDD" are often used interchangeably. ASD, in turn, is a subset of the broader autism phenotype, which describes individuals who may not have ASD but do have autistic-like traits, such as avoiding eye contact.
+
+Autism can also be divided into syndromal and non-syndromal autism; the syndromal autism is associated with severe or profound intellectual disability or a congenital syndrome with physical symptoms, such as tuberous sclerosis. Although individuals with Asperger syndrome tend to perform better cognitively than those with autism, the extent of the overlap between Asperger syndrome, HFA, and non-syndromal autism is unclear.
+
+Some studies have reported diagnoses of autism in children due to a loss of language or social skills, as opposed to a failure to make progress, typically from 15 to 30 months of age. The validity of this distinction remains controversial; it is possible that regressive autism is a specific subtype, or that there is a continuum of behaviors between autism with and without regression.
+
+Research into causes has been hampered by the inability to identify biologically meaningful subgroups within the autistic population and by the traditional boundaries between the disciplines of psychiatry, psychology, neurology and pediatrics. Newer technologies such as fMRI and diffusion tensor imaging can help identify biologically relevant phenotypes (observable traits) that can be viewed on brain scans, to help further neurogenetic studies of autism; one example is lowered activity in the fusiform face area of the brain, which is associated with impaired perception of people versus objects. It has been proposed to classify autism using genetics as well as behavior.
+
+Autism has long been thought to cover a wide spectrum, ranging from individuals with severe impairments—who may be silent, developmentally disabled, and prone to frequent repetitive behavior such as hand flapping and rocking—to high functioning individuals who may have active but distinctly odd social approaches, narrowly focused interests, and verbose, pedantic communication. Because the behavior spectrum is continuous, boundaries between diagnostic categories are necessarily somewhat arbitrary. Sometimes the syndrome is divided into low-, medium- or high-functioning autism (LFA, MFA, and HFA), based on IQ thresholds. Some people have called for an end to the terms "high-functioning" and "low-functioning" due to lack of nuance and the potential for a person's needs or abilities to be overlooked.
+
+About half of parents of children with ASD notice their child's unusual behaviors by age 18 months, and about four-fifths notice by age 24 months. According to an article, failure to meet any of the following milestones "is an absolute indication to proceed with further evaluations. Delay in referral for such testing may delay early diagnosis and treatment and affect the long-term outcome".
+
+The United States Preventive Services Task Force in 2016 found it was unclear if screening was beneficial or harmful among children in whom there is no concerns. The Japanese practice is to screen all children for ASD at 18 and 24 months, using autism-specific formal screening tests. In contrast, in the UK, children whose families or doctors recognize possible signs of autism are screened. It is not known which approach is more effective. Screening tools include the Modified Checklist for Autism in Toddlers (M-CHAT), the Early Screening of Autistic Traits Questionnaire, and the First Year Inventory; initial data on M-CHAT and its predecessor, the Checklist for Autism in Toddlers (CHAT), on children aged 18–30 months suggests that it is best used in a clinical setting and that it has low sensitivity (many false-negatives) but good specificity (few false-positives). It may be more accurate to precede these tests with a broadband screener that does not distinguish ASD from other developmental disorders. Screening tools designed for one culture's norms for behaviors like eye contact may be inappropriate for a different culture. Although genetic screening for autism is generally still impractical, it can be considered in some cases, such as children with neurological symptoms and dysmorphic features.
+
+While infection with rubella during pregnancy causes fewer than 1% of cases of autism, vaccination against rubella can prevent many of those cases.
+
+The main goals when treating children with autism are to lessen associated deficits and family distress, and to increase quality of life and functional independence. In general, higher IQs are correlated with greater responsiveness to treatment and improved treatment outcomes. No single treatment is best and treatment is typically tailored to the child's needs. Families and the educational system are the main resources for treatment. Services should be carried out by behavior analysts, special education teachers, speech pathologists, and licensed psychologists. Studies of interventions have methodological problems that prevent definitive conclusions about efficacy. However, the development of evidence-based interventions has advanced in recent years. Although many psychosocial interventions have some positive evidence, suggesting that some form of treatment is preferable to no treatment, the methodological quality of systematic reviews of these studies has generally been poor, their clinical results are mostly tentative, and there is little evidence for the relative effectiveness of treatment options. Intensive, sustained special education programs and behavior therapy early in life can help children acquire self-care, communication, and job skills, and often improve functioning and decrease symptom severity and maladaptive behaviors; claims that intervention by around age three years is crucial are not substantiated. While medications have not been found to help with core symptoms, they may be used for associated symptoms, such as irritability, inattention, or repetitive behavior patterns.
+
+Educational interventions often used include applied behavior analysis (ABA), developmental models, structured teaching, speech and language therapy, social skills therapy, and occupational therapy. Among these approaches, interventions either treat autistic features comprehensively, or focalize treatment on a specific area of deficit. The quality of research for early intensive behavioral intervention (EIBI)—a treatment procedure incorporating over thirty hours per week of the structured type of ABA that is carried out with very young children—is currently low, and more vigorous research designs with larger sample sizes are needed. Two theoretical frameworks outlined for early childhood intervention include structured and naturalistic ABA interventions, and developmental social pragmatic models (DSP). One interventional strategy utilizes a parent training model, which teaches parents how to implement various ABA and DSP techniques, allowing for parents to disseminate interventions themselves. Various DSP programs have been developed to explicitly deliver intervention systems through at-home parent implementation. Despite the recent development of parent training models, these interventions have demonstrated effectiveness in numerous studies, being evaluated as a probable efficacious mode of treatment.
+
+Early, intensive ABA therapy has demonstrated effectiveness in enhancing communication and adaptive functioning in preschool children; it is also well-established for improving the intellectual performance of that age group. Similarly, a teacher-implemented intervention that utilizes a more naturalistic form of ABA combined with a developmental social pragmatic approach has been found to be beneficial in improving social-communication skills in young children, although there is less evidence in its treatment of global symptoms. Neuropsychological reports are often poorly communicated to educators, resulting in a gap between what a report recommends and what education is provided. It is not known whether treatment programs for children lead to significant improvements after the children grow up, and the limited research on the effectiveness of adult residential programs shows mixed results. The appropriateness of including children with varying severity of autism spectrum disorders in the general education population is a subject of current debate among educators and researchers.
+
+Medications may be used to treat ASD symptoms that interfere with integrating a child into home or school when behavioral treatment fails. They may also be used for associated health problems, such as ADHD or anxiety. More than half of US children diagnosed with ASD are prescribed psychoactive drugs or anticonvulsants, with the most common drug classes being antidepressants, stimulants, and antipsychotics. The atypical antipsychotic drugs risperidone and aripiprazole are FDA-approved for treating associated aggressive and self-injurious behaviors. However, their side effects must be weighed against their potential benefits, and people with autism may respond atypically. Side effects, for example, may include weight gain, tiredness, drooling, and aggression. SSRI antidepressants, such as fluoxetine and fluvoxamine, have been shown to be effective in reducing repetitive and ritualistic behaviors, while the stimulant medication methylphenidate is beneficial for some children with co-morbid inattentiveness or hyperactivity. There is scant reliable research about the effectiveness or safety of drug treatments for adolescents and adults with ASD. No known medication relieves autism's core symptoms of social and communication impairments. Experiments in mice have reversed or reduced some symptoms related to autism by replacing or modulating gene function, suggesting the possibility of targeting therapies to specific rare mutations known to cause autism.
+
+Although many alternative therapies and interventions are available, few are supported by scientific studies. Treatment approaches have little empirical support in quality-of-life contexts, and many programs focus on success measures that lack predictive validity and real-world relevance. Some alternative treatments may place the child at risk. The preference that children with autism have for unconventional foods can lead to reduction in bone cortical thickness with this being greater in those on casein-free diets, as a consequence of the low intake of calcium and vitamin D; however, suboptimal bone development in ASD has also been associated with lack of exercise and gastrointestinal disorders. In 2005, botched chelation therapy killed a five-year-old child with autism. Chelation is not recommended for people with ASD since the associated risks outweigh any potential benefits. Another alternative medicine practice with no evidence is CEASE therapy, a mixture of homeopathy, supplements, and 'vaccine detoxing'.
+
+Although popularly used as an alternative treatment for people with autism, as of 2018 there is no good evidence to recommend a gluten- and casein-free diet as a standard treatment. A 2018 review concluded that it may be a therapeutic option for specific groups of children with autism, such as those with known food intolerances or allergies, or with food intolerance markers. The authors analyzed the prospective trials conducted to date that studied the efficacy of the gluten- and casein-free diet in children with ASD (4 in total). All of them compared gluten- and casein-free diet versus normal diet with a control group (2 double-blind randomized controlled trials, 1 double-blind crossover trial, 1 single-blind trial). In two of the studies, whose duration was 12 and 24 months, a significant improvement in ASD symptoms (efficacy rate 50%) was identified. In the other two studies, whose duration was 3 months, no significant effect was observed. The authors concluded that a longer duration of the diet may be necessary to achieve the improvement of the ASD symptoms. Other problems documented in the trials carried out include transgressions of the diet, small sample size, the heterogeneity of the participants and the possibility of a placebo effect.
+
+In the subset of people who have gluten sensitivity there is limited evidence that suggests that a gluten-free diet may improve some autistic behaviors.
+
+There is tentative evidence that music therapy may improve social interactions, verbal communication, and non-verbal communication skills. There has been early research looking at hyperbaric treatments in children with autism. Studies on pet therapy have shown positive effects.
+
+There is no known cure. The degree of symptoms can decrease, occasionally to the extent that people lose their diagnosis of ASD; this occurs sometimes after intensive treatment and sometimes not. It is not known how often recovery happens; reported rates in unselected samples have ranged from 3% to 25%. Most children with autism acquire language by age five or younger, though a few have developed communication skills in later years. Many children with autism lack social support, future employment opportunities or self-determination. Although core difficulties tend to persist, symptoms often become less severe with age.
+
+Few high-quality studies address long-term prognosis. Some adults show modest improvement in communication skills, but a few decline; no study has focused on autism after midlife. Acquiring language before age six, having an IQ above 50, and having a marketable skill all predict better outcomes; independent living is unlikely with severe autism.
+
+Many individuals with autism face significant obstacles in transitioning to adulthood. Compared to the general population individuals with autism are more likely to be unemployed and to have never had a job. About half of people in their 20s with autism are not employed.
+
+Most recent reviews tend to estimate a prevalence of 1–2 per 1,000 for autism and close to 6 per 1,000 for ASD as of 2007. A 2016 survey in the United States reported a rate of 25 per 1,000 children for ASD. Globally, autism affects an estimated 24.8 million people , while Asperger syndrome affects a further 37.2 million. In 2012, the NHS estimated that the overall prevalence of autism among adults aged 18 years and over in the UK was 1.1%. Rates of PDD-NOS's has been estimated at 3.7 per 1,000, Asperger syndrome at roughly 0.6 per 1,000, and childhood disintegrative disorder at 0.02 per 1,000. CDC estimates about 1 out of 59 (1.7%) for 2014, an increase from 1 out of every 68 children (1.5%) for 2010.
+
+The number of reported cases of autism increased dramatically in the 1990s and early 2000s. This increase is largely attributable to changes in diagnostic practices, referral patterns, availability of services, age at diagnosis, and public awareness, though unidentified environmental risk factors cannot be ruled out. The available evidence does not rule out the possibility that autism's true prevalence has increased; a real increase would suggest directing more attention and funding toward changing environmental factors instead of continuing to focus on genetics.
+
+Boys are at higher risk for ASD than girls. The sex ratio averages 4.3:1 and is greatly modified by cognitive impairment: it may be close to 2:1 with intellectual disability and more than 5.5:1 without. Several theories about the higher prevalence in males have been investigated, but the cause of the difference is unconfirmed; one theory is that females are underdiagnosed.
+
+Although the evidence does not implicate any single pregnancy-related risk factor as a cause of autism, the risk of autism is associated with advanced age in either parent, and with diabetes, bleeding, and use of psychiatric drugs in the mother during pregnancy. The risk is greater with older fathers than with older mothers; two potential explanations are the known increase in mutation burden in older sperm, and the hypothesis that men marry later if they carry genetic liability and show some signs of autism. Most professionals believe that race, ethnicity, and socioeconomic background do not affect the occurrence of autism.
+
+Several other conditions are common in children with autism. They include:
+
+A few examples of autistic symptoms and treatments were described long before autism was named. The "Table Talk" of Martin Luther, compiled by his notetaker, Mathesius, contains the story of a 12-year-old boy who may have been severely autistic. Luther reportedly thought the boy was a soulless mass of flesh possessed by the devil, and suggested that he be suffocated, although a later critic has cast doubt on the veracity of this report. The earliest well-documented case of autism is that of Hugh Blair of Borgue, as detailed in a 1747 court case in which his brother successfully petitioned to annul Blair's marriage to gain Blair's inheritance. The Wild Boy of Aveyron, a feral child caught in 1798, showed several signs of autism; the medical student Jean Itard treated him with a behavioral program designed to help him form social attachments and to induce speech via imitation.
+
+The New Latin word "autismus" (English translation "autism") was coined by the Swiss psychiatrist Eugen Bleuler in 1910 as he was defining symptoms of schizophrenia. He derived it from the Greek word "autós" (αὐτός, meaning "self"), and used it to mean morbid self-admiration, referring to "autistic withdrawal of the patient to his fantasies, against which any influence from outside becomes an intolerable disturbance". A Soviet child psychiatrist, Grunya Sukhareva, described a similar syndrome that was published in Russian in 1925, and in German in 1926.
+
+The word "autism" first took its modern sense in 1938 when Hans Asperger of the Vienna University Hospital adopted Bleuler's terminology "autistic psychopaths" in a lecture in German about child psychology. Asperger was investigating an ASD now known as Asperger syndrome, though for various reasons it was not widely recognized as a separate diagnosis until 1981. Leo Kanner of the Johns Hopkins Hospital first used "autism" in its modern sense in English when he introduced the label "early infantile autism" in a 1943 report of 11 children with striking behavioral similarities. Almost all the characteristics described in Kanner's first paper on the subject, notably "autistic aloneness" and "insistence on sameness", are still regarded as typical of the autistic spectrum of disorders. It is not known whether Kanner derived the term independently of Asperger.
+
+Donald Triplett was the first person diagnosed with autism. He was diagnosed by Kanner after being first examined in 1938, and was labeled as "case 1". Triplett was noted for his savant abilities, particularly being able to name musical notes played on a piano and to mentally multiply numbers. His father, Oliver, described him as socially withdrawn but interested in number patterns, music notes, letters of the alphabet, and U.S. president pictures. By the age of 2, he had the ability to recite the 23rd Psalm and memorized 25 questions and answers from the Presbyterian catechism. He was also interested in creating musical chords.
+
+Kanner's reuse of "autism" led to decades of confused terminology like "infantile schizophrenia", and child psychiatry's focus on maternal deprivation led to misconceptions of autism as an infant's response to "refrigerator mothers". Starting in the late 1960s autism was established as a separate syndrome.
+
+As late as the mid-1970s there was little evidence of a genetic role in autism; while in 2007 it was believed to be one of the most heritable psychiatric conditions. Although the rise of parent organizations and the destigmatization of childhood ASD have affected how ASD is viewed, parents continue to feel social stigma in situations where their child's autistic behavior is perceived negatively, and many primary care physicians and medical specialists express some beliefs consistent with outdated autism research.
+
+It took until 1980 for the DSM-III to differentiate autism from childhood schizophrenia. In 1987, the DSM-III-R provided a checklist for diagnosing autism. In May 2013, the DSM-5 was released, updating the classification for pervasive developmental disorders. The grouping of disorders, including PDD-NOS, autism, Asperger syndrome, Rett syndrome, and CDD, has been removed and replaced with the general term of Autism Spectrum Disorders. The two categories that exist are impaired social communication and/or interaction, and restricted and/or repetitive behaviors.
+
+The Internet has helped autistic individuals bypass nonverbal cues and emotional sharing that they find difficult to deal with, and has given them a way to form online communities and work remotely. Societal and cultural aspects of autism have developed: some in the community seek a cure, while others believe that autism is simply another way of being.
+
+An autistic culture has emerged, accompanied by the autistic rights and neurodiversity movements. Events include World Autism Awareness Day, Autism Sunday, Autistic Pride Day, Autreat, and others. Organizations dedicated to promoting awareness of autism include Autistic Self Advocacy Network, Aspies For Freedom, Autism National Committee, and Autism Society of America. At the same time, some organizations, including Autism Speaks, have been condemned by disability rights organizations for failing to support autistic people. Social-science scholars study those with autism in hopes to learn more about "autism as a culture, transcultural comparisons... and research on social movements." While most autistic individuals do not have savant skills, many have been successful in their fields.
+
+The autism rights movement is a social movement within the context of disability rights that emphasizes the concept of neurodiversity, viewing the autism spectrum as a result of natural variations in the human brain rather than a disorder to be cured. The autism rights movement advocates for including greater acceptance of autistic behaviors; therapies that focus on coping skills rather than on imitating the behaviors of those without autism, and the recognition of the autistic community as a minority group. Autism rights or neurodiversity advocates believe that the autism spectrum is genetic and should be accepted as a natural expression of the human genome. This perspective is distinct from two other likewise distinct views: the medical perspective, that autism is caused by a genetic defect and should be addressed by targeting the autism gene(s), and fringe theories that autism is caused by environmental factors such as vaccines. A common criticism against autistic activists is that the majority of them are "high-functioning" or have Asperger syndrome and do not represent the views of "low-functioning" autistic people. 
+
+About half of autistics are unemployed, and one third of those with graduate degrees may be unemployed. Among autistics who find work, most are employed in sheltered settings working for wages below the national minimum. While employers state hiring concerns about productivity and supervision, experienced employers of autistics give positive reports of above average memory and detail orientation as well as a high regard for rules and procedure in autistic employees. A majority of the economic burden of autism is caused by decreased earnings in the job market. Some studies also find decreased earning among parents who care for autistic children.
+
+
+</doc>
\ No newline at end of file

tests/test_data_collator.py

@@ -11,9 +11,11 @@ if is_torch_available():
         DataCollatorForLanguageModeling,
         DataCollatorForNextSentencePrediction,
         DataCollatorForPermutationLanguageModeling,
+        DataCollatorForSOP,
         GlueDataset,
         GlueDataTrainingArguments,
         LineByLineTextDataset,
+        LineByLineWithSOPTextDataset,
         TextDataset,
         TextDatasetForNextSentencePrediction,
         default_data_collator,
@@ -21,6 +23,7 @@ if is_torch_available():
 
 
 PATH_SAMPLE_TEXT = "./tests/fixtures/sample_text.txt"
+PATH_SAMPLE_TEXT_DIR = "./tests/fixtures/tests_samples/wiki_text"
 
 
 @require_torch
@@ -168,3 +171,19 @@ class DataCollatorIntegrationTest(unittest.TestCase):
         self.assertEqual(batch["token_type_ids"].shape, torch.Size((total_samples, 512)))
         self.assertEqual(batch["masked_lm_labels"].shape, torch.Size((total_samples, 512)))
         self.assertEqual(batch["next_sentence_label"].shape, torch.Size((total_samples,)))
+
+    def test_sop(self):
+        tokenizer = AutoTokenizer.from_pretrained("albert-base-v2")
+        data_collator = DataCollatorForSOP(tokenizer)
+
+        dataset = LineByLineWithSOPTextDataset(tokenizer, file_dir=PATH_SAMPLE_TEXT_DIR, block_size=512)
+        examples = [dataset[i] for i in range(len(dataset))]
+        batch = data_collator(examples)
+        self.assertIsInstance(batch, dict)
+
+        # Since there are randomly generated false samples, the total number of samples is not fixed.
+        total_samples = batch["input_ids"].shape[0]
+        self.assertEqual(batch["input_ids"].shape, torch.Size((total_samples, 512)))
+        self.assertEqual(batch["token_type_ids"].shape, torch.Size((total_samples, 512)))
+        self.assertEqual(batch["labels"].shape, torch.Size((total_samples, 512)))
+        self.assertEqual(batch["sentence_order_label"].shape, torch.Size((total_samples,)))