update issue numbers from jiant-dev to jiant transfer (#1196)

jiant/proj/main/modeling/taskmodels.py

@@ -46,7 +46,7 @@ def __init__(self, encoder, regression_head: heads.RegressionHead):
 
  def forward(self, batch, task, tokenizer, compute_loss: bool = False):
  encoder_output = get_output_from_encoder_and_batch(encoder=self.encoder, batch=batch)
- # TODO: Abuse of notation - these aren't really logits (Issue #45)
+ # TODO: Abuse of notation - these aren't really logits (issue #1187)
  logits = self.regression_head(pooled=encoder_output.pooled)
  if compute_loss:
  loss_fct = nn.MSELoss()
@@ -247,9 +247,9 @@ def forward(self, batch, task, tokenizer, compute_loss: bool = False):
  else:
  raise TypeError(type(self.pooler_head))
 
- # TODO: Abuse of notation - these aren't really logits (Issue #45)
+ # TODO: Abuse of notation - these aren't really logits (issue #1187)
  if compute_loss:
- # TODO: make this optional? (Issue #45)
+ # TODO: make this optional? (issue #1187)
  return LogitsAndLossOutput(
  logits=logits,
  loss=torch.tensor([0.0]), # This is a horrible hack

jiant/proj/main/preprocessing.py

@@ -39,7 +39,7 @@ def smart_truncate(dataset: torch_utils.ListDataset, max_seq_length: int, verbos
  range_idx = np.arange(max_seq_length)
  for datum in dataset.data:
  # TODO: document why reshape and max happen here (for cola this isn't necessary).
- # (Issue #47)
+ # (issue #1185)
  indexer = datum["data_row"].input_mask.reshape(-1, max_seq_length).max(-2)
  valid_length_ls.append(range_idx[indexer.astype(bool)].max() + 1)
  max_valid_length = max(valid_length_ls)
@@ -107,7 +107,7 @@ def convert_examples_to_dataset(
  Args:
  task (Task): Task object
  examples (list[Example]): list of task Examples.
- tokenizer: TODO (Issue #44)
+ tokenizer: TODO (issue #1188)
  feat_spec (FeaturizationSpec): Tokenization-related metadata.
  phase (str): string identifying the data subset (e.g., train, val or test).
  verbose: If True, display progress bar.
@@ -157,7 +157,7 @@ def tokenize_and_featurize(
  Args:
  task (Task): Task object
  examples (list[Example]): list of task Examples.
- tokenizer: TODO (Issue #44)
+ tokenizer: TODO (issue #1188)
  feat_spec (FeaturizationSpec): Tokenization-related metadata.
  phase (str): string identifying the data subset (e.g., train, val or test).
  verbose: If True, display progress bar.
@@ -166,7 +166,7 @@ def tokenize_and_featurize(
  List DataRows containing tokenized and featurized examples.
 
  """
- # TODO: Better solution (Issue #48)
+ # TODO: Better solution (issue #1184)
  if task.TASK_TYPE == TaskTypes.SQUAD_STYLE_QA:
  data_rows = []
  for example in maybe_tqdm(examples, desc="Tokenizing", verbose=verbose):
@@ -193,7 +193,7 @@ def iter_chunk_tokenize_and_featurize(
  Args:
  task (Task): Task object
  examples (list[Example]): list of task Examples.
- tokenizer: TODO (Issue #44)
+ tokenizer: TODO (issue #1188)
  feat_spec (FeaturizationSpec): Tokenization-related metadata.
  phase (str): string identifying the data subset (e.g., train, val or test).
  verbose: If True, display progress bar.
@@ -203,7 +203,7 @@ def iter_chunk_tokenize_and_featurize(
 
  """
  for example in maybe_tqdm(examples, desc="Tokenizing", verbose=verbose):
- # TODO: Better solution (Issue #48)
+ # TODO: Better solution (issue #1184)
  if task.TASK_TYPE == TaskTypes.SQUAD_STYLE_QA:
  yield from example.to_feature_list(
  tokenizer=tokenizer,

jiant/proj/main/runner.py

@@ -225,7 +225,7 @@ def complex_backpropagate(self, loss, gradient_accumulation_steps):
  )
 
  def get_runner_state(self):
- # TODO: Add fp16 (Issue #46)
+ # TODO: Add fp16 (issue #1186)
  state = {
  "model": torch_utils.get_model_for_saving(self.jiant_model).state_dict(),
  "optimizer": self.optimizer_scheduler.optimizer.state_dict(),

jiant/proj/main/tokenize_and_cache.py

@@ -39,7 +39,7 @@ def chunk_and_save(task, phase, examples, feat_spec, tokenizer, args: RunConfigu
  phase (str): string identifying the data subset (e.g., train, val or test).
  examples (list[Example]): list of task Examples.
  feat_spec: (FeaturizationSpec): Tokenization-related metadata.
- tokenizer: TODO (Issue #44)
+ tokenizer: TODO (issue #1188)
  args (RunConfiguration): run configuration object.
 
  """
@@ -71,7 +71,7 @@ def full_chunk_and_save(task, phase, examples, feat_spec, tokenizer, args: RunCo
  phase (str): string identifying the data subset (e.g., train, val or test).
  examples (list[Example]): list of task Examples.
  feat_spec: (FeaturizationSpec): Tokenization-related metadata.
- tokenizer: TODO (Issue #44)
+ tokenizer: TODO (issue #1188)
  args (RunConfiguration): run configuration object.
 
  """
@@ -108,7 +108,7 @@ def iter_chunk_and_save(task, phase, examples, feat_spec, tokenizer, args: RunCo
  phase (str): string identifying the data subset (e.g., train, val or test).
  examples (list[Example]): list of task Examples.
  feat_spec: (FeaturizationSpec): Tokenization-related metadata.
- tokenizer: TODO (Issue #44)
+ tokenizer: TODO (issue #1188)
  args (RunConfiguration): run configuration object.
 
  """

jiant/proj/simple/runscript.py

@@ -153,7 +153,7 @@ def run_simple(args: RunConfiguration):
  ),
  output_dir=os.path.join(args.exp_dir, "cache", task_name),
  phases=phases_to_do,
- # TODO: Need a strategy for task-specific max_seq_length issues (Issue #66)
+ # TODO: Need a strategy for task-specific max_seq_length issues (issue #1176)
  max_seq_length=args.max_seq_length,
  smart_truncate=True,
  do_iter=True,

jiant/scripts/benchmarks/xtreme/xtreme_submission.py

@@ -45,7 +45,7 @@ class RunConfiguration(zconf.RunConfig):
 
  # === Nuisance Parameters === #
  # Required for quickly setting up runner
- # Remove/refactor with config refactor (Issue #66)
+ # Remove/refactor with config refactor (issue #1176)
  learning_rate = zconf.attr(default=1e-5, type=float)
  adam_epsilon = zconf.attr(default=1e-8, type=float)
  max_grad_norm = zconf.attr(default=1.0, type=float)

jiant/shared/runner.py

@@ -31,7 +31,7 @@ def complex_backpropagate(
 def get_train_dataloader_from_cache(
  train_cache: caching.ChunkedFilesDataCache, task, train_batch_size: int
 ):
- # TODO: Expose buffer_size parameter (Issue #50)
+ # TODO: Expose buffer_size parameter (issue #1183)
  dataset = train_cache.get_iterable_dataset(buffer_size=10000, shuffle=True)
  train_dataloader = torch_utils.DataLoaderWithLength(
  dataset=dataset, batch_size=train_batch_size, collate_fn=task.collate_fn,

jiant/tasks/evaluate/core.py

@@ -908,7 +908,7 @@ def compute_metrics_from_preds_and_labels(cls, preds, labels):
 
 
 def get_evaluation_scheme_for_task(task) -> BaseEvaluationScheme:
- # TODO: move logic to task? (Issue #52)
+ # TODO: move logic to task? (issue #1182)
  if isinstance(
  task,
  (

jiant/tasks/lib/templates/hacky_tokenization_matching.py

@@ -1,4 +1,4 @@
-"""TODO: Remove when Tokenizers gets better (Issue #43)"""
+"""TODO: Remove when Tokenizers gets better (issue #1189)"""
 import transformers
 from jiant.tasks.utils import ExclusiveSpan
 
@@ -83,7 +83,7 @@ def roberta_flat_strip(tokens, return_indices=False):
 
 
 def xlm_roberta_flat_strip(tokens, return_indices=False):
- # TODO: Refactor to use general SentencePiece function (Issue #53)
+ # TODO: Refactor to use general SentencePiece function (issue #1181)
  return albert_flat_strip(tokens=tokens, return_indices=return_indices)
 
 

jiant/tasks/retrieval.py

@@ -147,7 +147,7 @@ def create_task_from_config(config: dict, base_path: Optional[str] = None, verbo
  task_class = get_task_class(config["task"])
  for k in config["paths"].keys():
  path = config["paths"][k]
- # TODO: Refactor paths (Issue #54)
+ # TODO: Refactor paths (issue #1180)
  if isinstance(path, str) and not os.path.isabs(path):
  assert base_path
  config["paths"][k] = os.path.join(base_path, path)

jiant/utils/torch_utils.py

@@ -94,7 +94,7 @@ def get_only_requires_grad(parameters, requires_grad=True):
  elif isinstance(parameters, dict):
  return {n: p for n, p in parameters if p.requires_grad == requires_grad}
  else:
- # TODO: Support generators (Issue #56)
+ # TODO: Support generators (issue #1178)
  raise RuntimeError("generators not yet supported")
 
 
@@ -111,7 +111,7 @@ def __getitem__(self, item):
 
 class DataLoaderWithLength(DataLoader):
  def __len__(self):
- # TODO: Revert after https://github.com/pytorch/pytorch/issues/36176 addressed  (Issue #55)
+ # TODO: Revert after https://github.com/pytorch/pytorch/issues/36176 addressed (issue #1179)
  # try:
  # return super().__len__()
  # except TypeError as e:

jiant/utils/zconf/core.py

@@ -58,7 +58,7 @@ def argparse_attr(
 
 
 def update_parser(parser, class_with_attributes: Any):
- # TODO: Write more checks/tests for the parser creation in general (Issue #57)
+ # TODO: Write more checks/tests for the parser creation in general (issue #1177)
  for attribute in class_with_attributes.__attrs_attrs__:
  if "argparse_kwargs" in attribute.metadata:
  argparse_kwargs = attribute.metadata["argparse_kwargs"]