transformers/tests/models/bart/test_modeling_flax_bart.py

# Copyright 2021 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import unittest

import numpy as np
import timeout_decorator  # noqa

from transformers import BartConfig, BartTokenizer, is_flax_available
from transformers.testing_utils import require_flax, slow

from ...generation.test_flax_utils import FlaxGenerationTesterMixin
from ...test_modeling_flax_common import FlaxModelTesterMixin, floats_tensor, ids_tensor, random_attention_mask


if is_flax_available():
    import os

    # The slow tests are often failing with OOM error on GPU
    # This makes JAX allocate exactly what is needed on demand, and deallocate memory that is no longer needed
    # but will be slower as stated here https://jax.readthedocs.io/en/latest/gpu_memory_allocation.html
    os.environ["XLA_PYTHON_CLIENT_ALLOCATOR"] = "platform"

    import jax
    import jax.numpy as jnp

    from transformers.models.bart.modeling_flax_bart import (
        FlaxBartForConditionalGeneration,
        FlaxBartForQuestionAnswering,
        FlaxBartForSequenceClassification,
        FlaxBartModel,
        shift_tokens_right,
    )


def prepare_bart_inputs_dict(
    config,
    input_ids,
    decoder_input_ids=None,
    attention_mask=None,
    decoder_attention_mask=None,
    head_mask=None,
    decoder_head_mask=None,
    cross_attn_head_mask=None,
):
    if attention_mask is None:
        attention_mask = np.where(input_ids != config.pad_token_id, 1, 0)
    if decoder_attention_mask is None:
        decoder_attention_mask = np.where(decoder_input_ids != config.pad_token_id, 1, 0)
    if head_mask is None:
        head_mask = np.ones((config.encoder_layers, config.encoder_attention_heads))
    if decoder_head_mask is None:
        decoder_head_mask = np.ones((config.decoder_layers, config.decoder_attention_heads))
    if cross_attn_head_mask is None:
        cross_attn_head_mask = np.ones((config.decoder_layers, config.decoder_attention_heads))
    return {
        "input_ids": input_ids,
        "decoder_input_ids": decoder_input_ids,
        "attention_mask": attention_mask,
        "decoder_attention_mask": attention_mask,
    }


class FlaxBartModelTester:
    def __init__(
        self,
        parent,
        batch_size=13,
        seq_length=7,
        is_training=True,
        use_labels=False,
        vocab_size=99,
        hidden_size=16,
        num_hidden_layers=2,
        num_attention_heads=4,
        intermediate_size=4,
        hidden_act="gelu",
        hidden_dropout_prob=0.1,
        attention_probs_dropout_prob=0.1,
        max_position_embeddings=32,
        eos_token_id=2,
        pad_token_id=1,
        bos_token_id=0,
        initializer_range=0.02,
    ):
        self.parent = parent
        self.batch_size = batch_size
        self.seq_length = seq_length
        self.is_training = is_training
        self.use_labels = use_labels
        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.intermediate_size = intermediate_size
        self.hidden_act = hidden_act
        self.hidden_dropout_prob = hidden_dropout_prob
        self.attention_probs_dropout_prob = attention_probs_dropout_prob
        self.max_position_embeddings = max_position_embeddings
        self.eos_token_id = eos_token_id
        self.pad_token_id = pad_token_id
        self.bos_token_id = bos_token_id
        self.initializer_range = initializer_range

    def prepare_config_and_inputs(self):
        input_ids = np.clip(ids_tensor([self.batch_size, self.seq_length - 1], self.vocab_size), 3, self.vocab_size)
        input_ids = np.concatenate((input_ids, 2 * np.ones((self.batch_size, 1), dtype=np.int64)), -1)

        decoder_input_ids = shift_tokens_right(input_ids, 1, 2)

        config = BartConfig(
            vocab_size=self.vocab_size,
            d_model=self.hidden_size,
            encoder_layers=self.num_hidden_layers,
            decoder_layers=self.num_hidden_layers,
            encoder_attention_heads=self.num_attention_heads,
            decoder_attention_heads=self.num_attention_heads,
            encoder_ffn_dim=self.intermediate_size,
            decoder_ffn_dim=self.intermediate_size,
            dropout=self.hidden_dropout_prob,
            attention_dropout=self.attention_probs_dropout_prob,
            max_position_embeddings=self.max_position_embeddings,
            eos_token_id=self.eos_token_id,
            bos_token_id=self.bos_token_id,
            pad_token_id=self.pad_token_id,
            initializer_range=self.initializer_range,
            use_cache=False,
        )
        inputs_dict = prepare_bart_inputs_dict(config, input_ids, decoder_input_ids)
        return config, inputs_dict

    def prepare_config_and_inputs_for_common(self):
        config, inputs_dict = self.prepare_config_and_inputs()
        return config, inputs_dict

    def check_use_cache_forward(self, model_class_name, config, inputs_dict):
        max_decoder_length = 20
        model = model_class_name(config)

        encoder_outputs = model.encode(inputs_dict["input_ids"])

        decoder_input_ids, decoder_attention_mask = (
            inputs_dict["decoder_input_ids"],
            inputs_dict["decoder_attention_mask"],
        )

        past_key_values = model.init_cache(decoder_input_ids.shape[0], max_decoder_length, encoder_outputs)
        decoder_attention_mask = jnp.ones((decoder_input_ids.shape[0], max_decoder_length), dtype="i4")

        decoder_position_ids = jnp.broadcast_to(
            jnp.arange(decoder_input_ids.shape[-1] - 1)[None, :],
            (decoder_input_ids.shape[0], decoder_input_ids.shape[-1] - 1),
        )
        outputs_cache = model.decode(
            decoder_input_ids[:, :-1],
            encoder_outputs,
            decoder_attention_mask=decoder_attention_mask,
            past_key_values=past_key_values,
            decoder_position_ids=decoder_position_ids,
        )

        decoder_position_ids = jnp.array(decoder_input_ids.shape[0] * [[decoder_input_ids.shape[-1] - 1]], dtype="i4")
        outputs_cache_next = model.decode(
            decoder_input_ids[:, -1:],
            encoder_outputs,
            decoder_attention_mask=decoder_attention_mask,
            past_key_values=outputs_cache.past_key_values,
            decoder_position_ids=decoder_position_ids,
        )

        outputs = model.decode(decoder_input_ids, encoder_outputs)

        diff = np.max(np.abs((outputs_cache_next[0][:, -1, :5] - outputs[0][:, -1, :5])))
        self.parent.assertTrue(diff < 1e-3, msg=f"Max diff is {diff}")

    def check_use_cache_forward_with_attn_mask(self, model_class_name, config, inputs_dict):
        max_decoder_length = 20
        model = model_class_name(config)

        encoder_outputs = model.encode(inputs_dict["input_ids"])

        decoder_input_ids, decoder_attention_mask = (
            inputs_dict["decoder_input_ids"],
            inputs_dict["decoder_attention_mask"],
        )

        decoder_attention_mask_cache = jnp.concatenate(
            [
                decoder_attention_mask,
                jnp.zeros((decoder_attention_mask.shape[0], max_decoder_length - decoder_attention_mask.shape[1])),
            ],
            axis=-1,
        )

        past_key_values = model.init_cache(decoder_input_ids.shape[0], max_decoder_length, encoder_outputs)
        decoder_position_ids = jnp.broadcast_to(
            jnp.arange(decoder_input_ids.shape[-1] - 1)[None, :],
            (decoder_input_ids.shape[0], decoder_input_ids.shape[-1] - 1),
        )

        outputs_cache = model.decode(
            decoder_input_ids[:, :-1],
            encoder_outputs,
            decoder_attention_mask=decoder_attention_mask_cache,
            past_key_values=past_key_values,
            decoder_position_ids=decoder_position_ids,
        )
        decoder_position_ids = jnp.array(decoder_input_ids.shape[0] * [[decoder_input_ids.shape[-1] - 1]], dtype="i4")
        outputs_cache_next = model.decode(
            decoder_input_ids[:, -1:],
            encoder_outputs,
            past_key_values=outputs_cache.past_key_values,
            decoder_attention_mask=decoder_attention_mask_cache,
            decoder_position_ids=decoder_position_ids,
        )

        outputs = model.decode(decoder_input_ids, encoder_outputs, decoder_attention_mask=decoder_attention_mask)

        diff = np.max(np.abs((outputs_cache_next[0][:, -1, :5] - outputs[0][:, -1, :5])))
        self.parent.assertTrue(diff < 1e-3, msg=f"Max diff is {diff}")


@require_flax
class BartHeadTests(unittest.TestCase):
    vocab_size = 99

    def _get_config_and_data(self):
        input_ids = np.array(
            [
                [71, 82, 18, 33, 46, 91, 2],
                [68, 34, 26, 58, 30, 82, 2],
                [5, 97, 17, 39, 94, 40, 2],
                [76, 83, 94, 25, 70, 78, 2],
                [87, 59, 41, 35, 48, 66, 2],
                [55, 13, 16, 58, 5, 2, 1],  # note padding
                [64, 27, 31, 51, 12, 75, 2],
                [52, 64, 86, 17, 83, 39, 2],
                [48, 61, 9, 24, 71, 82, 2],
                [26, 1, 60, 48, 22, 13, 2],
                [21, 5, 62, 28, 14, 76, 2],
                [45, 98, 37, 86, 59, 48, 2],
                [70, 70, 50, 9, 28, 0, 2],
            ],
            dtype=np.int64,
        )

        batch_size = input_ids.shape[0]
        config = BartConfig(
            vocab_size=self.vocab_size,
            d_model=24,
            encoder_layers=2,
            decoder_layers=2,
            encoder_attention_heads=2,
            decoder_attention_heads=2,
            encoder_ffn_dim=32,
            decoder_ffn_dim=32,
            max_position_embeddings=48,
            eos_token_id=2,
            pad_token_id=1,
            bos_token_id=0,
        )
        return config, input_ids, batch_size

    def test_sequence_classification_forward(self):
        config, input_ids, batch_size = self._get_config_and_data()
        model = FlaxBartForSequenceClassification(config)
        outputs = model(input_ids=input_ids, decoder_input_ids=input_ids)
        expected_shape = (batch_size, config.num_labels)
        self.assertEqual(outputs["logits"].shape, expected_shape)

    def test_question_answering_forward(self):
        config, input_ids, batch_size = self._get_config_and_data()
        model = FlaxBartForQuestionAnswering(config)
        outputs = model(input_ids=input_ids)

        self.assertEqual(outputs["start_logits"].shape, input_ids.shape)
        self.assertEqual(outputs["end_logits"].shape, input_ids.shape)

    # @timeout_decorator.timeout(1)  # not working with the decorator so far
    def test_lm_forward(self):
        config, input_ids, batch_size = self._get_config_and_data()
        lm_model = FlaxBartForConditionalGeneration(config)
        outputs = lm_model(input_ids=input_ids)
        expected_shape = (batch_size, input_ids.shape[1], config.vocab_size)
        self.assertEqual(outputs["logits"].shape, expected_shape)

    def test_lm_uneven_forward(self):
        config = BartConfig(
            vocab_size=self.vocab_size,
            d_model=14,
            encoder_layers=2,
            decoder_layers=2,
            encoder_attention_heads=2,
            decoder_attention_heads=2,
            encoder_ffn_dim=8,
            decoder_ffn_dim=8,
            max_position_embeddings=48,
        )
        lm_model = FlaxBartForConditionalGeneration(config)
        context = np.array([[71, 82, 18, 33, 46, 91, 2], [68, 34, 26, 58, 30, 2, 1]], dtype=np.int64)
        summary = np.array([[82, 71, 82, 18, 2], [58, 68, 2, 1, 1]], dtype=np.int64)
        outputs = lm_model(input_ids=context, decoder_input_ids=summary)
        expected_shape = (*summary.shape, config.vocab_size)
        self.assertEqual(outputs["logits"].shape, expected_shape)

    def test_shift_tokens_right(self):
        input_ids = np.array([[71, 82, 18, 33, 2, 1, 1], [68, 34, 26, 58, 30, 82, 2]], dtype=np.int64)
        shifted = shift_tokens_right(input_ids, 1, 2)
        n_pad_before = np.equal(input_ids, 1).astype(np.float32).sum()
        n_pad_after = np.equal(shifted, 1).astype(np.float32).sum()
        self.assertEqual(shifted.shape, input_ids.shape)
        self.assertEqual(n_pad_after, n_pad_before - 1)
        self.assertTrue(np.equal(shifted[:, 0], 2).all())


@require_flax
class FlaxBartModelTest(FlaxModelTesterMixin, unittest.TestCase, FlaxGenerationTesterMixin):
    is_encoder_decoder = True
    all_model_classes = (
        (
            FlaxBartModel,
            FlaxBartForConditionalGeneration,
            FlaxBartForSequenceClassification,
            FlaxBartForQuestionAnswering,
        )
        if is_flax_available()
        else ()
    )
    all_generative_model_classes = (FlaxBartForConditionalGeneration,) if is_flax_available() else ()

    def setUp(self):
        self.model_tester = FlaxBartModelTester(self)

    def test_use_cache_forward(self):
        config, inputs_dict = self.model_tester.prepare_config_and_inputs()
        for model_class in self.all_model_classes:
            self.model_tester.check_use_cache_forward(model_class, config, inputs_dict)

    def test_use_cache_forward_with_attn_mask(self):
        config, inputs_dict = self.model_tester.prepare_config_and_inputs()
        for model_class in self.all_model_classes:
            self.model_tester.check_use_cache_forward_with_attn_mask(model_class, config, inputs_dict)

    def test_encode(self):
        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()

        for model_class in self.all_model_classes:
            with self.subTest(model_class.__name__):
                prepared_inputs_dict = self._prepare_for_class(inputs_dict, model_class)
                model = model_class(config)

                @jax.jit
                def encode_jitted(input_ids, attention_mask=None, **kwargs):
                    return model.encode(input_ids=input_ids, attention_mask=attention_mask)

                with self.subTest("JIT Enabled"):
                    jitted_outputs = encode_jitted(**prepared_inputs_dict).to_tuple()

                with self.subTest("JIT Disabled"):
                    with jax.disable_jit():
                        outputs = encode_jitted(**prepared_inputs_dict).to_tuple()

                self.assertEqual(len(outputs), len(jitted_outputs))
                for jitted_output, output in zip(jitted_outputs, outputs):
                    self.assertEqual(jitted_output.shape, output.shape)

    def test_decode(self):
        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()

        for model_class in self.all_model_classes:
            with self.subTest(model_class.__name__):
                model = model_class(config)
                encoder_outputs = model.encode(inputs_dict["input_ids"], inputs_dict["attention_mask"])

                prepared_inputs_dict = {
                    "decoder_input_ids": inputs_dict["decoder_input_ids"],
                    "decoder_attention_mask": inputs_dict["decoder_attention_mask"],
                    "encoder_outputs": encoder_outputs,
                }

                @jax.jit
                def decode_jitted(decoder_input_ids, decoder_attention_mask, encoder_outputs):
                    return model.decode(
                        decoder_input_ids=decoder_input_ids,
                        decoder_attention_mask=decoder_attention_mask,
                        encoder_outputs=encoder_outputs,
                    )

                with self.subTest("JIT Enabled"):
                    jitted_outputs = decode_jitted(**prepared_inputs_dict).to_tuple()

                with self.subTest("JIT Disabled"):
                    with jax.disable_jit():
                        outputs = decode_jitted(**prepared_inputs_dict).to_tuple()

                self.assertEqual(len(outputs), len(jitted_outputs))
                for jitted_output, output in zip(jitted_outputs, outputs):
                    self.assertEqual(jitted_output.shape, output.shape)

    @slow
    def test_model_from_pretrained(self):
        for model_class_name in self.all_model_classes:
            model = model_class_name.from_pretrained("facebook/bart-base", from_pt=True)
            # FlaxBartForSequenceClassification expects eos token in input_ids
            input_ids = np.ones((1, 1)) * model.config.eos_token_id
            outputs = model(input_ids)
            self.assertIsNotNone(outputs)

    @slow
    def test_summarization_fast(self):
        model = FlaxBartForConditionalGeneration.from_pretrained("sshleifer/distilbart-cnn-6-6")
        tokenizer = BartTokenizer.from_pretrained("sshleifer/distilbart-cnn-6-6")

        input_str = (
            "This sentence is made of three parts. Each part is important on its own. One part is about animals, the"
            " other part about planes, and the last part about housing."
        )

        input_ids = tokenizer(input_str, return_tensors="np").input_ids
        sequences = model.generate(input_ids, num_beams=2, min_length=None, max_length=20).sequences

        output_str = tokenizer.batch_decode(sequences)[0]

        assert (
            output_str == "</s><s>This sentence is made of three parts. One part is about animals, the other part</s>"
        )

    @slow
    def test_cnn_summarization_same_as_fairseq(self):
        model = FlaxBartForConditionalGeneration.from_pretrained("facebook/bart-large-cnn")
        tokenizer = BartTokenizer.from_pretrained("facebook/bart-large-cnn")

        FRANCE_ARTICLE = (  # @noq
            " Marseille, France (CNN)The French prosecutor leading an investigation into the crash of Germanwings"
            " Flight 9525 insisted Wednesday that he was not aware of any video footage from on board the plane."
            ' Marseille prosecutor Brice Robin told CNN that "so far no videos were used in the crash investigation."'
            ' He added, "A person who has such a video needs to immediately give it to the investigators." Robin\'s'
            " comments follow claims by two magazines, German daily Bild and French Paris Match, of a cell phone video"
            " showing the harrowing final seconds from on board Germanwings Flight 9525 as it crashed into the French"
            " Alps. All 150 on board were killed. Paris Match and Bild reported that the video was recovered from a"
            " phone at the wreckage site. The two publications described the supposed video, but did not post it on"
            " their websites. The publications said that they watched the video, which was found by a source close to"
            " the investigation. \"One can hear cries of 'My God' in several languages,\" Paris Match reported."
            ' "Metallic banging can also be heard more than three times, perhaps of the pilot trying to open the'
            " cockpit door with a heavy object.  Towards the end, after a heavy shake, stronger than the others, the"
            ' screaming intensifies. Then nothing." "It is a very disturbing scene," said Julian Reichelt,'
            " editor-in-chief of Bild online. An official with France's accident investigation agency, the BEA, said"
            " the agency is not aware of any such video. Lt. Col. Jean-Marc Menichini, a French Gendarmerie spokesman"
            " in charge of communications on rescue efforts around the Germanwings crash site, told CNN that the"
            ' reports were "completely wrong" and "unwarranted." Cell phones have been collected at the site, he said,'
            ' but that they "hadn\'t been exploited yet." Menichini said he believed the cell phones would need to be'
            " sent to the Criminal Research Institute in Rosny sous-Bois, near Paris, in order to be analyzed by"
            " specialized technicians working hand-in-hand with investigators. But none of the cell phones found so"
            " far have been sent to the institute, Menichini said. Asked whether staff involved in the search could"
            ' have leaked a memory card to the media, Menichini answered with a categorical "no." Reichelt told "Erin'
            ' Burnett: Outfront" that he had watched the video and stood by the report, saying Bild and Paris Match'
            ' are "very confident" that the clip is real. He noted that investigators only revealed they\'d recovered'
            ' cell phones from the crash site after Bild and Paris Match published their reports. "That is something'
            " we did not know before. ... Overall we can say many things of the investigation weren't revealed by the"
            ' investigation at the beginning," he said. What was mental state of Germanwings co-pilot? German airline'
            " Lufthansa confirmed Tuesday that co-pilot Andreas Lubitz had battled depression years before he took the"
            " controls of Germanwings Flight 9525, which he's accused of deliberately crashing last week in the"
            ' French Alps. Lubitz told his Lufthansa flight training school in 2009 that he had a "previous episode of'
            ' severe depression," the airline said Tuesday. Email correspondence between Lubitz and the school'
            " discovered in an internal investigation, Lufthansa said, included medical documents he submitted in"
            " connection with resuming his flight training. The announcement indicates that Lufthansa, the parent"
            " company of Germanwings, knew of Lubitz's battle with depression, allowed him to continue training and"
            " ultimately put him in the cockpit. Lufthansa, whose CEO Carsten Spohr previously said Lubitz was 100%"
            ' fit to fly, described its statement Tuesday as a "swift and seamless clarification" and said it was'
            " sharing the information and documents -- including training and medical records -- with public"
            " prosecutors. Spohr traveled to the crash site Wednesday, where recovery teams have been working for the"
            " past week to recover human remains and plane debris scattered across a steep mountainside. He saw the"
            " crisis center set up in Seyne-les-Alpes, laid a wreath in the village of Le Vernet, closer to the crash"
            " site, where grieving families have left flowers at a simple stone memorial. Menichini told CNN late"
            " Tuesday that no visible human remains were left at the site but recovery teams would keep searching."
            " French President Francois Hollande, speaking Tuesday, said that it should be possible to identify all"
            " the victims using DNA analysis by the end of the week, sooner than authorities had previously suggested."
            " In the meantime, the recovery of the victims' personal belongings will start Wednesday, Menichini said."
            " Among those personal belongings could be more cell phones belonging to the 144 passengers and six crew"
            " on board. Check out the latest from our correspondents . The details about Lubitz's correspondence with"
            " the flight school during his training were among several developments as investigators continued to"
            " delve into what caused the crash and Lubitz's possible motive for downing the jet. A Lufthansa"
            " spokesperson told CNN on Tuesday that Lubitz had a valid medical certificate, had passed all his"
            ' examinations and "held all the licenses required." Earlier, a spokesman for the prosecutor\'s office in'
            " Dusseldorf, Christoph Kumpa, said medical records reveal Lubitz suffered from suicidal tendencies at"
            " some point before his aviation career and underwent psychotherapy before he got his pilot's license."
            " Kumpa emphasized there's no evidence suggesting Lubitz was suicidal or acting aggressively before the"
            " crash. Investigators are looking into whether Lubitz feared his medical condition would cause him to"
            " lose his pilot's license, a European government official briefed on the investigation told CNN on"
            ' Tuesday. While flying was "a big part of his life," the source said, it\'s only one theory being'
            " considered. Another source, a law enforcement official briefed on the investigation, also told CNN that"
            " authorities believe the primary motive for Lubitz to bring down the plane was that he feared he would"
            " not be allowed to fly because of his medical problems. Lubitz's girlfriend told investigators he had"
            " seen an eye doctor and a neuropsychologist, both of whom deemed him unfit to work recently and concluded"
            " he had psychological issues, the European government official said. But no matter what details emerge"
            " about his previous mental health struggles, there's more to the story, said Brian Russell, a forensic"
            ' psychologist. "Psychology can explain why somebody would turn rage inward on themselves about the fact'
            " that maybe they weren't going to keep doing their job and they're upset about that and so they're"
            ' suicidal," he said. "But there is no mental illness that explains why somebody then feels entitled to'
            " also take that rage and turn it outward on 149 other people who had nothing to do with the person's"
            ' problems." Germanwings crash compensation: What we know . Who was the captain of Germanwings Flight'
            " 9525? CNN's Margot Haddad reported from Marseille and Pamela Brown from Dusseldorf, while Laura"
            " Smith-Spark wrote from London. CNN's Frederik Pleitgen, Pamela Boykoff, Antonia Mortensen, Sandrine"
            " Amiel and Anna-Maja Rappard contributed to this report."
        )

        SHORTER_ARTICLE = (
            " (CNN)The Palestinian Authority officially became the 123rd member of the International Criminal Court on"
            " Wednesday, a step that gives the court jurisdiction over alleged crimes in Palestinian territories. The"
            " formal accession was marked with a ceremony at The Hague, in the Netherlands, where the court is based."
            " The Palestinians signed the ICC's founding Rome Statute in January, when they also accepted its"
            ' jurisdiction over alleged crimes committed "in the occupied Palestinian territory, including East'
            ' Jerusalem, since June 13, 2014." Later that month, the ICC opened a preliminary examination into the'
            " situation in Palestinian territories, paving the way for possible war crimes investigations against"
            " Israelis. As members of the court, Palestinians may be subject to counter-charges as well. Israel and"
            " the United States, neither of which is an ICC member, opposed the Palestinians' efforts to join the"
            " body. But Palestinian Foreign Minister Riad al-Malki, speaking at Wednesday's ceremony, said it was a"
            ' move toward greater justice. "As Palestine formally becomes a State Party to the Rome Statute today, the'
            ' world is also a step closer to ending a long era of impunity and injustice," he said, according to an'
            ' ICC news release. "Indeed, today brings us closer to our shared goals of justice and peace." Judge'
            " Kuniko Ozaki, a vice president of the ICC, said acceding to the treaty was just the first step for the"
            ' Palestinians. "As the Rome Statute today enters into force for the State of Palestine, Palestine'
            " acquires all the rights as well as responsibilities that come with being a State Party to the Statute."
            ' These are substantive commitments, which cannot be taken lightly," she said. Rights group Human Rights'
            ' Watch welcomed the development. "Governments seeking to penalize Palestine for joining the ICC should'
            " immediately end their pressure, and countries that support universal acceptance of the court's treaty"
            ' should speak out to welcome its membership," said Balkees Jarrah, international justice counsel for the'
            " group. \"What's objectionable is the attempts to undermine international justice, not Palestine's"
            ' decision to join a treaty to which over 100 countries around the world are members." In January, when'
            " the preliminary ICC examination was opened, Israeli Prime Minister Benjamin Netanyahu described it as an"
            ' outrage, saying the court was overstepping its boundaries. The United States also said it "strongly"'
            " disagreed with the court's decision. \"As we have said repeatedly, we do not believe that Palestine is a"
            ' state and therefore we do not believe that it is eligible to join the ICC," the State Department said in'
            ' a statement. It urged the warring sides to resolve their differences through direct negotiations. "We'
            ' will continue to oppose actions against Israel at the ICC as counterproductive to the cause of peace,"'
            " it said. But the ICC begs to differ with the definition of a state for its purposes and refers to the"
            ' territories as "Palestine." While a preliminary examination is not a formal investigation, it allows the'
            " court to review evidence and determine whether to investigate suspects on both sides. Prosecutor Fatou"
            ' Bensouda said her office would "conduct its analysis in full independence and impartiality." The war'
            " between Israel and Hamas militants in Gaza last summer left more than 2,000 people dead. The inquiry"
            " will include alleged war crimes committed since June. The International Criminal Court was set up in"
            " 2002 to prosecute genocide, crimes against humanity and war crimes. CNN's Vasco Cotovio, Kareem Khadder"
            " and Faith Karimi contributed to this report."
        )

        # The below article tests that we don't add any hypotheses outside of the top n_beams
        IRAN_ARTICLE = (
            " (CNN)The United States and its negotiating partners reached a very strong framework agreement with Iran"
            " in Lausanne, Switzerland, on Thursday that limits Iran's nuclear program in such a way as to effectively"
            " block it from building a nuclear weapon. Expect pushback anyway, if the recent past is any harbinger."
            " Just last month, in an attempt to head off such an agreement, House Speaker John Boehner invited Israeli"
            " Prime Minister Benjamin Netanyahu to preemptively blast it before Congress, and 47 senators sent a"
            " letter to the Iranian leadership warning them away from a deal. The debate that has already begun since"
            " the announcement of the new framework will likely result in more heat than light. It will not be helped"
            " by the gathering swirl of dubious assumptions and doubtful assertions. Let us address some of these: ."
            " The most misleading assertion, despite universal rejection by experts, is that the negotiations'"
            " objective at the outset was the total elimination of any nuclear program in Iran. That is the position"
            " of Netanyahu and his acolytes in the U.S. Congress. But that is not and never was the objective. If it"
            " had been, there would have been no Iranian team at the negotiating table. Rather, the objective has"
            " always been to structure an agreement or series of agreements so that Iran could not covertly develop a"
            " nuclear arsenal before the United States and its allies could respond. The new framework has exceeded"
            " expectations in achieving that goal. It would reduce Iran's low-enriched uranium stockpile, cut by"
            " two-thirds its number of installed centrifuges and implement a rigorous inspection regime. Another"
            " dubious assumption of opponents is that the Iranian nuclear program is a covert weapons program. Despite"
            " sharp accusations by some in the United States and its allies, Iran denies having such a program, and"
            " U.S. intelligence contends that Iran has not yet made the decision to build a nuclear weapon. Iran's"
            " continued cooperation with International Atomic Energy Agency inspections is further evidence on this"
            " point, and we'll know even more about Iran's program in the coming months and years because of the deal."
            " In fact, the inspections provisions that are part of this agreement are designed to protect against any"
            " covert action by the Iranians. What's more, the rhetoric of some members of Congress has implied that"
            " the negotiations have been between only the United States and Iran (i.e., the 47 senators' letter"
            " warning that a deal might be killed by Congress or a future president). This of course is not the case."
            " The talks were between Iran and the five permanent members of the U.N. Security Council (United States,"
            " United Kingdom, France, China and Russia) plus Germany, dubbed the P5+1. While the United States has"
            " played a leading role in the effort, it negotiated the terms alongside its partners. If the agreement"
            " reached by the P5+1 is rejected by Congress, it could result in an unraveling of the sanctions on Iran"
            " and threaten NATO cohesion in other areas. Another questionable assertion is that this agreement"
            " contains a sunset clause, after which Iran will be free to do as it pleases. Again, this is not the"
            " case. Some of the restrictions on Iran's nuclear activities, such as uranium enrichment, will be eased"
            " or eliminated over time, as long as 15 years. But most importantly, the framework agreement includes"
            " Iran's ratification of the Additional Protocol, which allows IAEA inspectors expanded access to nuclear"
            " sites both declared and nondeclared. This provision will be permanent. It does not sunset. Thus, going"
            " forward, if Iran decides to enrich uranium to weapons-grade levels, monitors will be able to detect such"
            " a move in a matter of days and alert the U.N. Security Council. Many in Congress have said that the"
            ' agreement should be a formal treaty requiring the Senate to "advise and consent." But the issue is not'
            " suited for a treaty. Treaties impose equivalent obligations on all signatories. For example, the New"
            " START treaty limits Russia and the United States to 1,550 deployed strategic warheads. But any agreement"
            " with Iran will not be so balanced.  The restrictions and obligations in the final framework agreement"
            " will be imposed almost exclusively on Iran. The P5+1 are obligated only to ease and eventually remove"
            " most but not all economic sanctions, which were imposed as leverage to gain this final deal. Finally"
            " some insist that any agreement must address Iranian missile programs, human rights violations or support"
            " for Hamas or Hezbollah.  As important as these issues are, and they must indeed be addressed, they are"
            " unrelated to the most important aim of a nuclear deal: preventing a nuclear Iran.  To include them in"
            " the negotiations would be a poison pill. This agreement should be judged on its merits and on how it"
            " affects the security of our negotiating partners and allies, including Israel. Those judgments should be"
            " fact-based, not based on questionable assertions or dubious assumptions."
        )

        ARTICLE_SUBWAY = (
            " New York (CNN)When Liana Barrientos was 23 years old, she got married in Westchester County, New York. A"
            " year later, she got married again in Westchester County, but to a different man and without divorcing"
            " her first husband.  Only 18 days after that marriage, she got hitched yet again. Then, Barrientos"
            ' declared "I do" five more times, sometimes only within two weeks of each other. In 2010, she married'
            " once more, this time in the Bronx. In an application for a marriage license, she stated it was her"
            ' "first and only" marriage. Barrientos, now 39, is facing two criminal counts of "offering a false'
            ' instrument for filing in the first degree," referring to her false statements on the 2010 marriage'
            " license application, according to court documents. Prosecutors said the marriages were part of an"
            " immigration scam. On Friday, she pleaded not guilty at State Supreme Court in the Bronx, according to"
            " her attorney, Christopher Wright, who declined to comment further. After leaving court, Barrientos was"
            " arrested and charged with theft of service and criminal trespass for allegedly sneaking into the New"
            " York subway through an emergency exit, said Detective Annette Markowski, a police spokeswoman. In total,"
            " Barrientos has been married 10 times, with nine of her marriages occurring between 1999 and 2002.  All"
            " occurred either in Westchester County, Long Island, New Jersey or the Bronx. She is believed to still be"
            " married to four men, and at one time, she was married to eight men at once, prosecutors say. Prosecutors"
            " said the immigration scam involved some of her husbands, who filed for permanent residence status"
            " shortly after the marriages.  Any divorces happened only after such filings were approved. It was"
            " unclear whether any of the men will be prosecuted. The case was referred to the Bronx District"
            " Attorney's Office by Immigration and Customs Enforcement and the Department of Homeland Security's"
            ' Investigation Division. Seven of the men are from so-called "red-flagged" countries, including Egypt,'
            " Turkey, Georgia, Pakistan and Mali. Her eighth husband, Rashid Rajput, was deported in 2006 to his"
            " native Pakistan after an investigation by the Joint Terrorism Task Force. If convicted, Barrientos faces"
            " up to four years in prison.  Her next court appearance is scheduled for May 18."
        )

        dct = tokenizer.batch_encode_plus(
            [FRANCE_ARTICLE, SHORTER_ARTICLE, IRAN_ARTICLE, ARTICLE_SUBWAY],
            max_length=1024,
            padding="max_length",
            truncation_strategy="only_first",
            truncation=True,
            return_tensors="np",
        )

        self.assertEqual(1024, dct["input_ids"].shape[1])
        hypotheses_batch = model.generate(
            input_ids=dct["input_ids"],
            attention_mask=dct["attention_mask"],
            num_beams=2,
        ).sequences
        assert (hypotheses_batch[:, 1] == 0).all().item()

        EXPECTED = [
            "A French prosecutor says he is not aware of any video footage from on board the plane. Two German"
            " magazines claim to have found a cell phone video showing the crash. The publications say they watched"
            " the video, which was found by a source close to the investigation. All 150 on board the Germanwings"
            " flight were killed.",
            "Palestinian Authority becomes 123rd member of the International Criminal Court. The move gives the court"
            " jurisdiction over alleged crimes in Palestinian territories. Israel and the United States opposed the"
            " Palestinians' efforts to join the body. But Palestinian Foreign Minister Riad al-Malki said it was a"
            " move toward greater justice.",
            "U.S. and its negotiating partners reached a strong framework agreement with Iran. Peter Bergen: The"
            " debate that has already begun will likely result in more heat than light. Bergen: The most misleading"
            " assertion is that the negotiations' objective at the outset was the total elimination of any nuclear"
            " program.",
            "Liana Barrientos, 39, has been married 10 times, sometimes within two weeks of each other. Prosecutors"
            " say the marriages were part of an immigration scam. She pleaded not guilty at State Supreme Court in the"
            " Bronx on Friday. If convicted, Barrientos faces up to four years in prison.",
        ]

        generated_summaries = tokenizer.batch_decode(
            hypotheses_batch.tolist(), clean_up_tokenization_spaces=True, skip_special_tokens=True
        )
        assert generated_summaries == EXPECTED


class FlaxBartStandaloneDecoderModelTester:
    def __init__(
        self,
        parent,
        batch_size=13,
        seq_length=7,
        is_training=True,
        use_attention_mask=True,
        use_labels=False,
        vocab_size=99,
        hidden_size=16,
        num_hidden_layers=2,
        num_attention_heads=4,
        intermediate_size=4,
        hidden_act="gelu",
        hidden_dropout_prob=0.1,
        attention_probs_dropout_prob=0.1,
        max_position_embeddings=32,
        eos_token_id=2,
        pad_token_id=1,
        bos_token_id=0,
        initializer_range=0.02,
    ):
        self.parent = parent
        self.batch_size = batch_size
        self.seq_length = seq_length
        self.is_training = is_training
        self.use_attention_mask = use_attention_mask
        self.use_labels = use_labels
        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.intermediate_size = intermediate_size
        self.hidden_act = hidden_act
        self.hidden_dropout_prob = hidden_dropout_prob
        self.attention_probs_dropout_prob = attention_probs_dropout_prob
        self.max_position_embeddings = max_position_embeddings
        self.eos_token_id = eos_token_id
        self.pad_token_id = pad_token_id
        self.bos_token_id = bos_token_id
        self.initializer_range = initializer_range

    def prepare_config_and_inputs(self):
        input_ids = jnp.clip(ids_tensor([self.batch_size, self.seq_length], self.vocab_size), 3, self.vocab_size)

        attention_mask = None
        if self.use_attention_mask:
            attention_mask = random_attention_mask([self.batch_size, self.seq_length])

        config = BartConfig(
            vocab_size=self.vocab_size,
            d_model=self.hidden_size,
            encoder_layers=self.num_hidden_layers,
            decoder_layers=self.num_hidden_layers,
            encoder_attention_heads=self.num_attention_heads,
            decoder_attention_heads=self.num_attention_heads,
            encoder_ffn_dim=self.intermediate_size,
            decoder_ffn_dim=self.intermediate_size,
            dropout=self.hidden_dropout_prob,
            attention_dropout=self.attention_probs_dropout_prob,
            max_position_embeddings=self.max_position_embeddings,
            eos_token_id=self.eos_token_id,
            bos_token_id=self.bos_token_id,
            pad_token_id=self.pad_token_id,
            initializer_range=self.initializer_range,
            use_cache=False,
        )

        return config, input_ids, attention_mask

    def prepare_config_and_inputs_for_common(self):
        config_and_inputs = self.prepare_config_and_inputs()
        config, input_ids, attention_mask = config_and_inputs
        inputs_dict = {"input_ids": input_ids, "attention_mask": attention_mask}
        return config, inputs_dict

    def prepare_config_and_inputs_for_decoder(self):
        config, input_ids, attention_mask = self.prepare_config_and_inputs()

        encoder_hidden_states = floats_tensor([self.batch_size, self.seq_length, self.hidden_size])
        encoder_attention_mask = ids_tensor([self.batch_size, self.seq_length], vocab_size=2)

        return (
            config,
            input_ids,
            attention_mask,
            encoder_hidden_states,
            encoder_attention_mask,
        )