perf: collator with padding for tpu

bsmetadata/experiments/with_metadata.py

@@ -1,9 +1,10 @@
 import functools
 import logging
 
+from accelerate import DistributedType
 from datasets import config, load_dataset
 from torch.utils.data import DataLoader
-from transformers import default_data_collator
+from transformers import DataCollatorWithPadding, default_data_collator
 
 from bsmetadata.metadata_utils import add_metadata_and_chunk_examples
 
@@ -152,15 +153,18 @@ def create_labels_column(examples):
     logger.info(f"  Num validation examples = {len(val_dataset)}")
 
     # DataLoaders creation:
+    data_collator = default_data_collator
+    if args.distributed_type == DistributedType.TPU:
+        data_collator = DataCollatorWithPadding(tokenizer, padding="max_length", max_length=args.max_seq_len)
     train_dataloader = DataLoader(
         train_dataset,
         shuffle=True,
-        collate_fn=default_data_collator,
+        collate_fn=data_collator,
         batch_size=args.per_device_train_batch_size,
     )
     val_dataloader1 = DataLoader(
         val_dataset,
-        collate_fn=default_data_collator,
+        collate_fn=data_collator,
         batch_size=args.per_device_eval_batch_size,
     )
     return train_dataloader, {"val1": val_dataloader1}

bsmetadata/experiments/without_metadata.py

@@ -1,8 +1,9 @@
 import logging
 
+from accelerate import DistributedType
 from datasets import config, load_dataset
 from torch.utils.data import DataLoader
-from transformers import default_data_collator
+from transformers import DataCollatorWithPadding, default_data_collator
 
 
 logger = logging.getLogger(__name__)
@@ -107,7 +108,7 @@ def get_dataloaders(tokenizer, args):
     text_column_name = "text" if "text" in column_names else column_names[0]
 
     def tokenize_function(examples):
-        return tokenizer(examples[text_column_name])
+        return tokenizer(examples[text_column_name], truncation=True, max_length=args.max_seq_len)
 
     logger.info("Tokenize dataset")
     tokenized_datasets = datasets.map(
@@ -179,15 +180,18 @@ def group_texts(examples):
     logger.info(f"  Num validation examples = {len(val_dataset)}")
 
     # DataLoaders creation:
+    data_collator = default_data_collator
+    if args.distributed_type == DistributedType.TPU:
+        data_collator = DataCollatorWithPadding(tokenizer, padding="max_length", max_length=args.max_seq_len)
     train_dataloader = DataLoader(
         train_dataset,
         shuffle=True,
-        collate_fn=default_data_collator,
+        collate_fn=data_collator,
         batch_size=args.per_device_train_batch_size,
     )
     val_dataloader1 = DataLoader(
         val_dataset,
-        collate_fn=default_data_collator,
+        collate_fn=data_collator,
         batch_size=args.per_device_eval_batch_size,
     )
     return train_dataloader, {"val1": val_dataloader1}

bsmetadata/input_pipeline.py

@@ -1,6 +1,8 @@
 from dataclasses import dataclass, field
 from typing import Optional
 
+from accelerate import DistributedType
+
 from bsmetadata.metadata_utils import MetadataConfig
 
 
@@ -53,6 +55,7 @@ class DataConfig:
             " the dataset. If you are using `with_metadata` the recommended batch size is 1.."
         },
     )
+    distributed_type: DistributedType = field(default=DistributedType.NO)
 
 
 def get_dataloaders(tokenizer, cfg: DataConfig):

bsmetadata/metadata_utils.py

@@ -71,7 +71,7 @@ def add_metadata_and_chunk_examples(
             # Get the global metadata prefix that is prepended to each training example.
             metadata_prefix = create_metadata_prefix(example, cfg)
             metadata_prefix_encoded = (
-                tokenizer.encode_plus(cfg.metadata_prefix_start_seq + metadata_prefix).input_ids
+                tokenizer.encode_plus(cfg.metadata_prefix_start_seq + metadata_prefix, truncation=True).input_ids
                 if metadata_prefix
                 else []
             )
@@ -89,7 +89,7 @@ def add_metadata_and_chunk_examples(
             text_with_local_metadata = " " + text_with_local_metadata
             char_level_metadata_mask = [False] + char_level_metadata_mask
 
-        text_with_local_metadata_encoded = tokenizer.encode_plus(text_with_local_metadata)
+        text_with_local_metadata_encoded = tokenizer.encode_plus(text_with_local_metadata, truncation=True)
 
         def is_metadata(idx: int) -> bool:
             char_span = text_with_local_metadata_encoded.token_to_chars(idx)

bsmetadata/train.py

@@ -136,6 +136,9 @@ def loss_fn(batch, outputs, metadata_mask=None):
 
 @hydra.main(config_path=None, config_name="config")
 def main(args: CFG) -> None:
+    accelerator = Accelerator()
+    args.data_config.distributed_type = accelerator.distributed_type
+
     print(OmegaConf.to_yaml(args))
     config_dict = OmegaConf.to_container(args)
 
@@ -144,7 +147,6 @@ def main(args: CFG) -> None:
     args = OmegaConf.to_object(args)
 
     set_seed(args.seed)
-    accelerator = Accelerator()
     is_local_main_process = accelerator.is_local_main_process
     tqdm = partial(original_tqdm, disable=not is_local_main_process, position=0)