Merge branch 'anmolg/validation_1' into 'main'

Validation dataset update 1

See merge request ADLR/megatron-lm!455

megatron/arguments.py

@@ -838,12 +838,31 @@ def _add_data_args(parser):
                        help='Path to the training dataset. Accepted format:'
                        '1) a single data path, 2) multiple datasets in the'
                        'form: dataset1-weight dataset1-path dataset2-weight '
-                       'dataset2-path ...')
+                       'dataset2-path ... It is used with --split when a '
+                       'single dataset used for all three: train, valid '
+                       'and test. It is exclusive to the other '
+                       '--*-data-path args')
     group.add_argument('--split', type=str, default='969, 30, 1',
                        help='Comma-separated list of proportions for training,'
                        ' validation, and test split. For example the split '
                        '`90,5,5` will use 90%% of data for training, 5%% for '
                        'validation and 5%% for test.')
+    group.add_argument('--train-data-path', nargs='*', default=None,
+                       help='Path to the training dataset. Accepted format:'
+                       '1) a single data path, 2) multiple datasets in the'
+                       'form: dataset1-weight dataset1-path dataset2-weight '
+                       'dataset2-path ...')
+    group.add_argument('--valid-data-path', nargs='*', default=None,
+                       help='Path to the validation dataset. Accepted format:'
+                       '1) a single data path, 2) multiple datasets in the'
+                       'form: dataset1-weight dataset1-path dataset2-weight '
+                       'dataset2-path ...')
+    group.add_argument('--test-data-path', nargs='*', default=None,
+                       help='Path to the test dataset. Accepted format:'
+                       '1) a single data path, 2) multiple datasets in the'
+                       'form: dataset1-weight dataset1-path dataset2-weight '
+                       'dataset2-path ...')
+
     group.add_argument('--vocab-file', type=str, default=None,
                        help='Path to the vocab file.')
     group.add_argument('--merge-file', type=str, default=None,

megatron/data/dataset_utils.py

@@ -63,12 +63,18 @@ def get_datasets_weights_and_num_samples(data_prefix,
     # Add 0.5% (the 1.005 factor) so in case the bleding dataset does
     # not uniformly distribute the number of samples, we still have
     # samples left to feed to the network.
-    datasets_train_valid_test_num_samples = []
-    for weight in weights:
-        datasets_train_valid_test_num_samples.append(
-            [int(math.ceil(val * weight * 1.005))
-             for val in train_valid_test_num_samples])
-
+    if isinstance(train_valid_test_num_samples, list):
+        datasets_train_valid_test_num_samples = []
+        for weight in weights:
+            datasets_train_valid_test_num_samples.append(
+                [int(math.ceil(val * weight * 1.005))
+                for val in train_valid_test_num_samples])
+    else:
+        # Used when separate dataset files are provided for train,
+        # valid and test
+        datasets_train_valid_test_num_samples = [
+            int(math.ceil(train_valid_test_num_samples * weight * 1.005))
+            for weight in weights]
 
     return prefixes, weights, datasets_train_valid_test_num_samples
 

megatron/data/gpt_dataset.py

@@ -15,53 +15,134 @@ from megatron.data.dataset_utils import get_train_valid_test_split_
 from megatron.data.indexed_dataset import make_dataset as make_indexed_dataset
 
 
-def build_train_valid_test_datasets(data_prefix, data_impl, splits_string,
-                                    train_valid_test_num_samples,
-                                    seq_length, seed, skip_warmup):
+def build_train_valid_test_datasets(data_prefix, data_impl,
+                                    splits_string, train_valid_test_num_samples,
+                                    seq_length, seed, skip_warmup,
+                                    train_data_prefix=None, valid_data_prefix=None,
+                                    test_data_prefix=None,):
     """Build train, valid, and test datasets."""
 
-    # Single dataset.
+    if data_prefix:
+        print_rank_0("Single data path provided for train, valid & test")
+        # Single dataset.
+        if len(data_prefix) == 1:
+            return _build_train_valid_test_datasets(data_prefix[0],
+                                                    data_impl, splits_string,
+                                                    train_valid_test_num_samples,
+                                                    seq_length, seed, skip_warmup)
+
+        # Blending dataset.
+        # Parse the values.
+        output = get_datasets_weights_and_num_samples(data_prefix,
+                                                    train_valid_test_num_samples)
+        prefixes, weights, datasets_train_valid_test_num_samples = output
+
+        # Build individual datasets.
+        train_datasets = []
+        valid_datasets = []
+        test_datasets = []
+        for i in range(len(prefixes)):
+            train_ds, valid_ds, test_ds = _build_train_valid_test_datasets(
+                prefixes[i], data_impl, splits_string,
+                datasets_train_valid_test_num_samples[i],
+                seq_length, seed, skip_warmup)
+            if train_ds:
+                train_datasets.append(train_ds)
+            if valid_ds:
+                valid_datasets.append(valid_ds)
+            if test_ds:
+                test_datasets.append(test_ds)
+
+        # Blend.
+        blending_train_dataset = None
+        if train_datasets:
+            blending_train_dataset = BlendableDataset(train_datasets, weights)
+        blending_valid_dataset = None
+        if valid_datasets:
+            blending_valid_dataset = BlendableDataset(valid_datasets, weights)
+        blending_test_dataset = None
+        if test_datasets:
+            blending_test_dataset = BlendableDataset(test_datasets, weights)
+
+        return (blending_train_dataset, blending_valid_dataset,
+                blending_test_dataset)
+    else:
+        print_rank_0("Separate data paths provided for train, valid & test. Split string will be ignored.")
+
+        train_dataset, valid_dataset, test_dataset = None, None, None
+        # Single dataset.
+        if train_data_prefix is not None:
+            train_dataset = build_dataset("train", train_data_prefix, data_impl,
+                                        train_valid_test_num_samples[0], seq_length, seed,
+                                        skip_warmup)
+
+        if valid_data_prefix is not None:
+            valid_dataset = build_dataset("valid", valid_data_prefix, data_impl,
+                                    train_valid_test_num_samples[1], seq_length, seed,
+                                    False)
+
+        if test_data_prefix is not None:
+            test_dataset = build_dataset("test", test_data_prefix, data_impl,
+                                    train_valid_test_num_samples[2], seq_length, seed,
+                                    False)
+
+        return (train_dataset, valid_dataset, test_dataset)
+
+
+def build_dataset(dataset_name, data_prefix, data_impl, num_samples, seq_length, seed, skip_warmup):
+    dataset = None
     if len(data_prefix) == 1:
-        return _build_train_valid_test_datasets(data_prefix[0],
-                                                data_impl, splits_string,
-                                                train_valid_test_num_samples,
-                                                seq_length, seed, skip_warmup)
-
-    # Blending dataset.
-    # Parse the values.
-    output = get_datasets_weights_and_num_samples(data_prefix,
-                                                  train_valid_test_num_samples)
-    prefixes, weights, datasets_train_valid_test_num_samples = output
-
-    # Build individual datasets.
-    train_datasets = []
-    valid_datasets = []
-    test_datasets = []
-    for i in range(len(prefixes)):
-        train_ds, valid_ds, test_ds = _build_train_valid_test_datasets(
-            prefixes[i], data_impl, splits_string,
-            datasets_train_valid_test_num_samples[i],
-            seq_length, seed, skip_warmup)
-        if train_ds:
-            train_datasets.append(train_ds)
-        if valid_ds:
-            valid_datasets.append(valid_ds)
-        if test_ds:
-            test_datasets.append(test_ds)
-
-    # Blend.
-    blending_train_dataset = None
-    if train_datasets:
-        blending_train_dataset = BlendableDataset(train_datasets, weights)
-    blending_valid_dataset = None
-    if valid_datasets:
-        blending_valid_dataset = BlendableDataset(valid_datasets, weights)
-    blending_test_dataset = None
-    if test_datasets:
-        blending_test_dataset = BlendableDataset(test_datasets, weights)
-
-    return (blending_train_dataset, blending_valid_dataset,
-            blending_test_dataset)
+        dataset = _build_dataset(dataset_name,
+                        data_prefix[0], data_impl,
+                        num_samples, seq_length,
+                        seed, skip_warmup)
+    else:
+        # Blending dataset.
+        # Parse the values.
+        output = get_datasets_weights_and_num_samples(data_prefix, num_samples)
+        prefixes, weights, dataset_num_samples = output
+
+        # Build individual datasets.
+        datasets = []
+        for i in range(len(prefixes)):
+            ds = _build_dataset(dataset_name, prefixes[i],
+                            data_impl, dataset_num_samples[i],
+                            seq_length, seed, skip_warmup)
+            if ds:
+                datasets.append(ds)
+
+        if datasets:
+            dataset = BlendableDataset(datasets, weights)
+
+    return dataset
+
+
+def _build_dataset(dataset_name, data_prefix, data_impl,
+                num_samples, seq_length, seed, skip_warmup):
+    """
+    Build dataset. This method is called when individual
+    train, valid, test datasets are provided
+    """
+
+    # Indexed dataset.
+    indexed_dataset = get_indexed_dataset_(data_prefix,
+                                           data_impl,
+                                           skip_warmup)
+
+    total_num_of_documents = indexed_dataset.sizes.shape[0]
+
+    print_rank_0('    {}:'.format(dataset_name))
+    print_rank_0('     document indices in [0, {}) total of {} '
+                 'documents'.format(total_num_of_documents, total_num_of_documents))
+
+    documents = np.arange(start=0, stop=total_num_of_documents,
+                        step=1, dtype=np.int32)
+
+    dataset = GPTDataset(dataset_name, data_prefix,
+                        documents, indexed_dataset,
+                        num_samples, seq_length, seed)
+
+    return dataset
 
 
 def _build_train_valid_test_datasets(data_prefix, data_impl, splits_string,

pretrain_gpt.py

@@ -100,7 +100,10 @@ def train_valid_test_datasets_provider(train_val_test_num_samples):
         train_valid_test_num_samples=train_val_test_num_samples,
         seq_length=args.seq_length,
         seed=args.seed,
-        skip_warmup=(not args.mmap_warmup))
+        skip_warmup=(not args.mmap_warmup),
+        train_data_prefix=args.train_data_path,
+        valid_data_prefix=args.valid_data_path,
+        test_data_prefix=args.test_data_path,)
     print_rank_0("> finished creating GPT datasets ...")
 
     return train_ds, valid_ds, test_ds