Add support for pretraining recurring span selection to Splinter (#17247)

* Add SplinterForSpanSelection for pre-training recurring span selection.

* Formatting.

* Rename SplinterForSpanSelection to SplinterForPreTraining.

* Ensure repo consistency

* Fixup changes

* Address SplinterForPreTraining PR comments

* Incorporate feedback and derive multiple question tokens per example.

* Update src/transformers/models/splinter/modeling_splinter.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update src/transformers/models/splinter/modeling_splinter.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Jean Vancoppenole <jean.vancoppenolle@retresco.de>
Co-authored-by: Tobias Günther <tobias.guenther@retresco.de>
Co-authored-by: Tobias Günther <github@tobigue.de>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

docs/source/en/model_doc/splinter.mdx

@@ -72,3 +72,8 @@ This model was contributed by [yuvalkirstain](https://huggingface.co/yuvalkirsta
 
 [[autodoc]] SplinterForQuestionAnswering
     - forward
+
+## SplinterForPreTraining
+
+[[autodoc]] SplinterForPreTraining
+    - forward

src/transformers/__init__.py

@@ -1532,6 +1532,7 @@ else:
     _import_structure["models.splinter"].extend(
         [
             "SPLINTER_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "SplinterForPreTraining",
             "SplinterForQuestionAnswering",
             "SplinterLayer",
             "SplinterModel",
@@ -3830,6 +3831,7 @@ if TYPE_CHECKING:
         from .models.speech_to_text_2 import Speech2Text2ForCausalLM, Speech2Text2PreTrainedModel
         from .models.splinter import (
             SPLINTER_PRETRAINED_MODEL_ARCHIVE_LIST,
+            SplinterForPreTraining,
             SplinterForQuestionAnswering,
             SplinterLayer,
             SplinterModel,

src/transformers/models/auto/modeling_auto.py

@@ -161,6 +161,7 @@ MODEL_FOR_PRETRAINING_MAPPING_NAMES = OrderedDict(
         ("openai-gpt", "OpenAIGPTLMHeadModel"),
         ("retribert", "RetriBertModel"),
         ("roberta", "RobertaForMaskedLM"),
+        ("splinter", "SplinterForPreTraining"),
         ("squeezebert", "SqueezeBertForMaskedLM"),
         ("t5", "T5ForConditionalGeneration"),
         ("tapas", "TapasForMaskedLM"),

src/transformers/models/splinter/__init__.py

@@ -42,6 +42,7 @@ else:
     _import_structure["modeling_splinter"] = [
         "SPLINTER_PRETRAINED_MODEL_ARCHIVE_LIST",
         "SplinterForQuestionAnswering",
+        "SplinterForPreTraining",
         "SplinterLayer",
         "SplinterModel",
         "SplinterPreTrainedModel",
@@ -68,6 +69,7 @@ if TYPE_CHECKING:
     else:
         from .modeling_splinter import (
             SPLINTER_PRETRAINED_MODEL_ARCHIVE_LIST,
+            SplinterForPreTraining,
             SplinterForQuestionAnswering,
             SplinterLayer,
             SplinterModel,

src/transformers/models/splinter/modeling_splinter.py

@@ -16,6 +16,7 @@
 
 
 import math
+from dataclasses import dataclass
 from typing import List, Optional, Tuple, Union
 
 import torch
@@ -24,7 +25,7 @@ from torch import nn
 from torch.nn import CrossEntropyLoss
 
 from ...activations import ACT2FN
-from ...modeling_outputs import BaseModelOutputWithPastAndCrossAttentions, QuestionAnsweringModelOutput
+from ...modeling_outputs import BaseModelOutputWithPastAndCrossAttentions, ModelOutput, QuestionAnsweringModelOutput
 from ...modeling_utils import PreTrainedModel
 from ...pytorch_utils import apply_chunking_to_forward, find_pruneable_heads_and_indices, prune_linear_layer
 from ...utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, logging
@@ -940,3 +941,171 @@ class SplinterForQuestionAnswering(SplinterPreTrainedModel):
             hidden_states=outputs.hidden_states,
             attentions=outputs.attentions,
         )
+
+
+@dataclass
+class SplinterForPreTrainingOutput(ModelOutput):
+    """
+    Class for outputs of Splinter as a span selection model.
+
+    Args:
+        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when start and end positions are provided):
+            Total span extraction loss is the sum of a Cross-Entropy for the start and end positions.
+        start_logits (`torch.FloatTensor` of shape `(batch_size, num_questions, sequence_length)`):
+            Span-start scores (before SoftMax).
+        end_logits (`torch.FloatTensor` of shape `(batch_size, num_questions, sequence_length)`):
+            Span-end scores (before SoftMax).
+        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
+            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
+            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
+
+            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
+        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
+            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
+            sequence_length)`.
+
+            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
+            heads.
+    """
+
+    loss: Optional[torch.FloatTensor] = None
+    start_logits: torch.FloatTensor = None
+    end_logits: torch.FloatTensor = None
+    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
+    attentions: Optional[Tuple[torch.FloatTensor]] = None
+
+
+@add_start_docstrings(
+    """
+    Splinter Model for the recurring span selection task as done during the pretraining. The difference to the QA task
+    is that we do not have a question, but multiple question tokens that replace the occurrences of recurring spans
+    instead.
+    """,
+    SPLINTER_START_DOCSTRING,
+)
+class SplinterForPreTraining(SplinterPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.splinter = SplinterModel(config)
+        self.splinter_qass = QuestionAwareSpanSelectionHead(config)
+        self.question_token_id = config.question_token_id
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(
+        SPLINTER_INPUTS_DOCSTRING.format("batch_size, num_questions, sequence_length")
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        token_type_ids: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        start_positions: Optional[torch.LongTensor] = None,
+        end_positions: Optional[torch.LongTensor] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        question_positions: Optional[torch.LongTensor] = None,
+    ) -> Union[Tuple, SplinterForPreTrainingOutput]:
+        r"""
+        start_positions (`torch.LongTensor` of shape `(batch_size, num_questions)`, *optional*):
+            Labels for position (index) of the start of the labelled span for computing the token classification loss.
+            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
+            are not taken into account for computing the loss.
+        end_positions (`torch.LongTensor` of shape `(batch_size, num_questions)`, *optional*):
+            Labels for position (index) of the end of the labelled span for computing the token classification loss.
+            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
+            are not taken into account for computing the loss.
+        question_positions (`torch.LongTensor` of shape `(batch_size, num_questions)`, *optional*):
+            The positions of all question tokens. If given, start_logits and end_logits will be of shape `(batch_size,
+            num_questions, sequence_length)`. If None, the first question token in each sequence in the batch will be
+            the only one for which start_logits and end_logits are calculated and they will be of shape `(batch_size,
+            sequence_length)`.
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if question_positions is None and start_positions is not None and end_positions is not None:
+            raise TypeError("question_positions must be specified in order to calculate the loss")
+
+        elif question_positions is None and input_ids is None:
+            raise TypeError("question_positions must be specified when input_embeds is used")
+
+        elif question_positions is None:
+            question_positions = self._prepare_question_positions(input_ids)
+
+        outputs = self.splinter(
+            input_ids,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        sequence_output = outputs[0]
+        batch_size, sequence_length, dim = sequence_output.size()
+        # [batch_size, num_questions, sequence_length]
+        start_logits, end_logits = self.splinter_qass(sequence_output, question_positions)
+
+        num_questions = question_positions.size(1)
+        if attention_mask is not None:
+            attention_mask_for_each_question = attention_mask.unsqueeze(1).expand(
+                batch_size, num_questions, sequence_length
+            )
+            start_logits = start_logits + (1 - attention_mask_for_each_question) * -10000.0
+            end_logits = end_logits + (1 - attention_mask_for_each_question) * -10000.0
+
+        total_loss = None
+        # [batch_size, num_questions, sequence_length]
+        if start_positions is not None and end_positions is not None:
+            # sometimes the start/end positions are outside our model inputs, we ignore these terms
+            start_positions.clamp_(0, max(0, sequence_length - 1))
+            end_positions.clamp_(0, max(0, sequence_length - 1))
+
+            # Ignore zero positions in the loss. Splinter never predicts zero
+            # during pretraining and zero is used for padding question
+            # tokens as well as for start and end positions of padded
+            # question tokens.
+            loss_fct = CrossEntropyLoss(ignore_index=self.config.pad_token_id)
+            start_loss = loss_fct(
+                start_logits.view(batch_size * num_questions, sequence_length),
+                start_positions.view(batch_size * num_questions),
+            )
+            end_loss = loss_fct(
+                end_logits.view(batch_size * num_questions, sequence_length),
+                end_positions.view(batch_size * num_questions),
+            )
+            total_loss = (start_loss + end_loss) / 2
+
+        if not return_dict:
+            output = (start_logits, end_logits) + outputs[1:]
+            return ((total_loss,) + output) if total_loss is not None else output
+
+        return SplinterForPreTrainingOutput(
+            loss=total_loss,
+            start_logits=start_logits,
+            end_logits=end_logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def _prepare_question_positions(self, input_ids: torch.Tensor) -> torch.Tensor:
+        rows, flat_positions = torch.where(input_ids == self.config.question_token_id)
+        num_questions = torch.bincount(rows)
+        positions = torch.full(
+            (input_ids.size(0), num_questions.max()),
+            self.config.pad_token_id,
+            dtype=torch.long,
+            device=input_ids.device,
+        )
+        cols = torch.cat([torch.arange(n) for n in num_questions])
+        positions[rows, cols] = flat_positions
+        return positions

tests/models/splinter/test_modeling_splinter.py

@@ -14,7 +14,7 @@
 # limitations under the License.
 """ Testing suite for the PyTorch Splinter model. """
 
-
+import copy
 import unittest
 
 from transformers import is_torch_available
@@ -27,7 +27,7 @@ from ...test_modeling_common import ModelTesterMixin, ids_tensor, random_attenti
 if is_torch_available():
     import torch
 
-    from transformers import SplinterConfig, SplinterForQuestionAnswering, SplinterModel
+    from transformers import SplinterConfig, SplinterForPreTraining, SplinterForQuestionAnswering, SplinterModel
     from transformers.models.splinter.modeling_splinter import SPLINTER_PRETRAINED_MODEL_ARCHIVE_LIST
 
 
@@ -36,6 +36,7 @@ class SplinterModelTester:
         self,
         parent,
         batch_size=13,
+        num_questions=3,
         seq_length=7,
         is_training=True,
         use_input_mask=True,
@@ -43,6 +44,7 @@ class SplinterModelTester:
         use_labels=True,
         vocab_size=99,
         hidden_size=32,
+        question_token_id=1,
         num_hidden_layers=5,
         num_attention_heads=4,
         intermediate_size=37,
@@ -59,6 +61,7 @@ class SplinterModelTester:
     ):
         self.parent = parent
         self.batch_size = batch_size
+        self.num_questions = num_questions
         self.seq_length = seq_length
         self.is_training = is_training
         self.use_input_mask = use_input_mask
@@ -66,6 +69,7 @@ class SplinterModelTester:
         self.use_labels = use_labels
         self.vocab_size = vocab_size
         self.hidden_size = hidden_size
+        self.question_token_id = question_token_id
         self.num_hidden_layers = num_hidden_layers
         self.num_attention_heads = num_attention_heads
         self.intermediate_size = intermediate_size
@@ -82,6 +86,7 @@ class SplinterModelTester:
 
     def prepare_config_and_inputs(self):
         input_ids = ids_tensor([self.batch_size, self.seq_length], self.vocab_size)
+        input_ids[:, 1] = self.question_token_id
 
         input_mask = None
         if self.use_input_mask:
@@ -91,13 +96,13 @@ class SplinterModelTester:
         if self.use_token_type_ids:
             token_type_ids = ids_tensor([self.batch_size, self.seq_length], self.type_vocab_size)
 
-        sequence_labels = None
-        token_labels = None
-        choice_labels = None
+        start_positions = None
+        end_positions = None
+        question_positions = None
         if self.use_labels:
-            sequence_labels = ids_tensor([self.batch_size], self.type_sequence_label_size)
-            token_labels = ids_tensor([self.batch_size, self.seq_length], self.num_labels)
-            choice_labels = ids_tensor([self.batch_size], self.num_choices)
+            start_positions = ids_tensor([self.batch_size, self.num_questions], self.type_sequence_label_size)
+            end_positions = ids_tensor([self.batch_size, self.num_questions], self.type_sequence_label_size)
+            question_positions = ids_tensor([self.batch_size, self.num_questions], self.num_labels)
 
         config = SplinterConfig(
             vocab_size=self.vocab_size,
@@ -112,12 +117,20 @@ class SplinterModelTester:
             type_vocab_size=self.type_vocab_size,
             is_decoder=False,
             initializer_range=self.initializer_range,
+            question_token_id=self.question_token_id,
         )
 
-        return config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+        return (config, input_ids, token_type_ids, input_mask, start_positions, end_positions, question_positions)
 
     def create_and_check_model(
-        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+        start_positions,
+        end_positions,
+        question_positions,
     ):
         model = SplinterModel(config=config)
         model.to(torch_device)
@@ -128,7 +141,14 @@ class SplinterModelTester:
         self.parent.assertEqual(result.last_hidden_state.shape, (self.batch_size, self.seq_length, self.hidden_size))
 
     def create_and_check_for_question_answering(
-        self, config, input_ids, token_type_ids, input_mask, sequence_labels, token_labels, choice_labels
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+        start_positions,
+        end_positions,
+        question_positions,
     ):
         model = SplinterForQuestionAnswering(config=config)
         model.to(torch_device)
@@ -137,12 +157,36 @@ class SplinterModelTester:
             input_ids,
             attention_mask=input_mask,
             token_type_ids=token_type_ids,
-            start_positions=sequence_labels,
-            end_positions=sequence_labels,
+            start_positions=start_positions[:, 0],
+            end_positions=end_positions[:, 0],
         )
         self.parent.assertEqual(result.start_logits.shape, (self.batch_size, self.seq_length))
         self.parent.assertEqual(result.end_logits.shape, (self.batch_size, self.seq_length))
 
+    def create_and_check_for_pretraining(
+        self,
+        config,
+        input_ids,
+        token_type_ids,
+        input_mask,
+        start_positions,
+        end_positions,
+        question_positions,
+    ):
+        model = SplinterForPreTraining(config=config)
+        model.to(torch_device)
+        model.eval()
+        result = model(
+            input_ids,
+            attention_mask=input_mask,
+            token_type_ids=token_type_ids,
+            start_positions=start_positions,
+            end_positions=end_positions,
+            question_positions=question_positions,
+        )
+        self.parent.assertEqual(result.start_logits.shape, (self.batch_size, self.num_questions, self.seq_length))
+        self.parent.assertEqual(result.end_logits.shape, (self.batch_size, self.num_questions, self.seq_length))
+
     def prepare_config_and_inputs_for_common(self):
         config_and_inputs = self.prepare_config_and_inputs()
         (
@@ -150,11 +194,15 @@ class SplinterModelTester:
             input_ids,
             token_type_ids,
             input_mask,
-            sequence_labels,
-            token_labels,
-            choice_labels,
+            start_positions,
+            end_positions,
+            question_positions,
         ) = config_and_inputs
-        inputs_dict = {"input_ids": input_ids, "token_type_ids": token_type_ids, "attention_mask": input_mask}
+        inputs_dict = {
+            "input_ids": input_ids,
+            "token_type_ids": token_type_ids,
+            "attention_mask": input_mask,
+        }
         return config, inputs_dict
 
 
@@ -165,11 +213,44 @@ class SplinterModelTest(ModelTesterMixin, unittest.TestCase):
         (
             SplinterModel,
             SplinterForQuestionAnswering,
+            SplinterForPreTraining,
         )
         if is_torch_available()
         else ()
     )
 
+    def _prepare_for_class(self, inputs_dict, model_class, return_labels=False):
+        inputs_dict = copy.deepcopy(inputs_dict)
+        if return_labels:
+            if issubclass(model_class, SplinterForPreTraining):
+                inputs_dict["start_positions"] = torch.zeros(
+                    self.model_tester.batch_size,
+                    self.model_tester.num_questions,
+                    dtype=torch.long,
+                    device=torch_device,
+                )
+                inputs_dict["end_positions"] = torch.zeros(
+                    self.model_tester.batch_size,
+                    self.model_tester.num_questions,
+                    dtype=torch.long,
+                    device=torch_device,
+                )
+                inputs_dict["question_positions"] = torch.zeros(
+                    self.model_tester.batch_size,
+                    self.model_tester.num_questions,
+                    dtype=torch.long,
+                    device=torch_device,
+                )
+            elif issubclass(model_class, SplinterForQuestionAnswering):
+                inputs_dict["start_positions"] = torch.zeros(
+                    self.model_tester.batch_size, dtype=torch.long, device=torch_device
+                )
+                inputs_dict["end_positions"] = torch.zeros(
+                    self.model_tester.batch_size, dtype=torch.long, device=torch_device
+                )
+
+        return inputs_dict
+
     def setUp(self):
         self.model_tester = SplinterModelTester(self)
         self.config_tester = ConfigTester(self, config_class=SplinterConfig, hidden_size=37)
@@ -191,6 +272,44 @@ class SplinterModelTest(ModelTesterMixin, unittest.TestCase):
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_for_question_answering(*config_and_inputs)
 
+    def test_for_pretraining(self):
+        config_and_inputs = self.model_tester.prepare_config_and_inputs()
+        self.model_tester.create_and_check_for_pretraining(*config_and_inputs)
+
+    def test_inputs_embeds(self):
+        config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+
+        for model_class in self.all_model_classes:
+            model = model_class(config)
+            model.to(torch_device)
+            model.eval()
+
+            inputs = copy.deepcopy(self._prepare_for_class(inputs_dict, model_class))
+
+            if not self.is_encoder_decoder:
+                input_ids = inputs["input_ids"]
+                del inputs["input_ids"]
+            else:
+                encoder_input_ids = inputs["input_ids"]
+                decoder_input_ids = inputs.get("decoder_input_ids", encoder_input_ids)
+                del inputs["input_ids"]
+                inputs.pop("decoder_input_ids", None)
+
+            wte = model.get_input_embeddings()
+            if not self.is_encoder_decoder:
+                inputs["inputs_embeds"] = wte(input_ids)
+            else:
+                inputs["inputs_embeds"] = wte(encoder_input_ids)
+                inputs["decoder_inputs_embeds"] = wte(decoder_input_ids)
+
+            with torch.no_grad():
+                if isinstance(model, SplinterForPreTraining):
+                    with self.assertRaises(TypeError):
+                        # question_positions must not be None.
+                        model(**inputs)[0]
+                else:
+                    model(**inputs)[0]
+
     @slow
     def test_model_from_pretrained(self):
         for model_name in SPLINTER_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
@@ -217,3 +336,122 @@ class SplinterModelIntegrationTest(unittest.TestCase):
 
         self.assertEqual(torch.argmax(output.start_logits), 10)
         self.assertEqual(torch.argmax(output.end_logits), 12)
+
+    @slow
+    def test_splinter_pretraining(self):
+        model = SplinterForPreTraining.from_pretrained("tau/splinter-base-qass")
+
+        # Input: "[CLS] [QUESTION] was born in [QUESTION] . Brad returned to the United Kingdom later . [SEP]"
+        # Output should be the spans "Brad" and "the United Kingdom"
+        input_ids = torch.tensor(
+            [[101, 104, 1108, 1255, 1107, 104, 119, 7796, 1608, 1106, 1103, 1244, 2325, 1224, 119, 102]]
+        )
+        question_positions = torch.tensor([[1, 5]], dtype=torch.long)
+        output = model(input_ids, question_positions=question_positions)
+
+        expected_shape = torch.Size((1, 2, 16))
+        self.assertEqual(output.start_logits.shape, expected_shape)
+        self.assertEqual(output.end_logits.shape, expected_shape)
+
+        self.assertEqual(torch.argmax(output.start_logits[0, 0]), 7)
+        self.assertEqual(torch.argmax(output.end_logits[0, 0]), 7)
+        self.assertEqual(torch.argmax(output.start_logits[0, 1]), 10)
+        self.assertEqual(torch.argmax(output.end_logits[0, 1]), 12)
+
+    @slow
+    def test_splinter_pretraining_loss_requires_question_positions(self):
+        model = SplinterForPreTraining.from_pretrained("tau/splinter-base-qass")
+
+        # Input: "[CLS] [QUESTION] was born in [QUESTION] . Brad returned to the United Kingdom later . [SEP]"
+        # Output should be the spans "Brad" and "the United Kingdom"
+        input_ids = torch.tensor(
+            [[101, 104, 1108, 1255, 1107, 104, 119, 7796, 1608, 1106, 1103, 1244, 2325, 1224, 119, 102]]
+        )
+        start_positions = torch.tensor([[7, 10]], dtype=torch.long)
+        end_positions = torch.tensor([7, 12], dtype=torch.long)
+        with self.assertRaises(TypeError):
+            model(
+                input_ids,
+                start_positions=start_positions,
+                end_positions=end_positions,
+            )
+
+    @slow
+    def test_splinter_pretraining_loss(self):
+        model = SplinterForPreTraining.from_pretrained("tau/splinter-base-qass")
+
+        # Input: "[CLS] [QUESTION] was born in [QUESTION] . Brad returned to the United Kingdom later . [SEP]"
+        # Output should be the spans "Brad" and "the United Kingdom"
+        input_ids = torch.tensor(
+            [
+                [101, 104, 1108, 1255, 1107, 104, 119, 7796, 1608, 1106, 1103, 1244, 2325, 1224, 119, 102],
+                [101, 104, 1108, 1255, 1107, 104, 119, 7796, 1608, 1106, 1103, 1244, 2325, 1224, 119, 102],
+            ]
+        )
+        start_positions = torch.tensor([[7, 10], [7, 10]], dtype=torch.long)
+        end_positions = torch.tensor([[7, 12], [7, 12]], dtype=torch.long)
+        question_positions = torch.tensor([[1, 5], [1, 5]], dtype=torch.long)
+        output = model(
+            input_ids,
+            start_positions=start_positions,
+            end_positions=end_positions,
+            question_positions=question_positions,
+        )
+        self.assertAlmostEqual(output.loss.item(), 0.0024, 4)
+
+    @slow
+    def test_splinter_pretraining_loss_with_padding(self):
+        model = SplinterForPreTraining.from_pretrained("tau/splinter-base-qass")
+
+        # Input: "[CLS] [QUESTION] was born in [QUESTION] . Brad returned to the United Kingdom later . [SEP]"
+        # Output should be the spans "Brad" and "the United Kingdom"
+        input_ids = torch.tensor(
+            [
+                [101, 104, 1108, 1255, 1107, 104, 119, 7796, 1608, 1106, 1103, 1244, 2325, 1224, 119, 102],
+            ]
+        )
+        start_positions = torch.tensor([[7, 10]], dtype=torch.long)
+        end_positions = torch.tensor([7, 12], dtype=torch.long)
+        question_positions = torch.tensor([[1, 5]], dtype=torch.long)
+        start_positions_with_padding = torch.tensor([[7, 10, 0]], dtype=torch.long)
+        end_positions_with_padding = torch.tensor([7, 12, 0], dtype=torch.long)
+        question_positions_with_padding = torch.tensor([[1, 5, 0]], dtype=torch.long)
+        output = model(
+            input_ids,
+            start_positions=start_positions,
+            end_positions=end_positions,
+            question_positions=question_positions,
+        )
+        output_with_padding = model(
+            input_ids,
+            start_positions=start_positions_with_padding,
+            end_positions=end_positions_with_padding,
+            question_positions=question_positions_with_padding,
+        )
+
+        self.assertAlmostEqual(output.loss.item(), output_with_padding.loss.item(), 4)
+
+        # Note that the original code uses 0 to denote padded question tokens
+        # and their start and end positions. As the pad_token_id of the model's
+        # config is used for the losse's ignore_index in SplinterForPreTraining,
+        # we add this test to ensure anybody making changes to the default
+        # value of the config, will be aware of the implication.
+        self.assertEqual(model.config.pad_token_id, 0)
+
+    @slow
+    def test_splinter_pretraining_prepare_question_positions(self):
+        model = SplinterForPreTraining.from_pretrained("tau/splinter-base-qass")
+
+        input_ids = torch.tensor(
+            [
+                [101, 104, 1, 2, 104, 3, 4, 102],
+                [101, 1, 104, 2, 104, 3, 104, 102],
+                [101, 1, 2, 104, 104, 3, 4, 102],
+                [101, 1, 2, 3, 4, 5, 104, 102],
+            ]
+        )
+        question_positions = torch.tensor([[1, 4, 0], [2, 4, 6], [3, 4, 0], [6, 0, 0]], dtype=torch.long)
+        output_without_positions = model(input_ids)
+        output_with_positions = model(input_ids, question_positions=question_positions)
+        self.assertTrue((output_without_positions.start_logits == output_with_positions.start_logits).all())
+        self.assertTrue((output_without_positions.end_logits == output_with_positions.end_logits).all())