Improvements for semantic deduplication and DAPT tutorial

config/sem_dedup_config.yaml

@@ -3,22 +3,23 @@ cache_dir: "semdedup_cache"
 num_files: 16
 
 # Embeddings configuration
-embeddings_save_loc: "embeddings"
 embedding_model_name_or_path: "sentence-transformers/all-MiniLM-L6-v2"
 embedding_batch_size: 128
+embeddings_save_loc: "embeddings"
+embedding_pooling_strategy: "mean_pooling"
+embedding_column: "embeddings"
 write_embeddings_to_disk: true
+write_to_filename: false
 
 # Clustering configuration
-clustering_save_loc: "clustering_results"
-n_clusters: 1000
-seed: 1234
 max_iter: 100
-kmeans_with_cos_dist: false
-
-# Semdedup configuration
-which_to_keep: "hard"
-largest_cluster_size_to_process: 100000
+n_clusters: 1000
+clustering_save_loc: "clustering_results"
 sim_metric: "cosine"
+which_to_keep: "hard"
+sort_clusters: true
+kmeans_with_cos_dist: false
+partition_size: "2gb"
 
 # Extract dedup configuration
 eps_thresholds:

docs/user-guide/semdedup.rst

@@ -42,22 +42,23 @@ Semantic deduplication in NeMo Curator can be configured using a YAML file. Here
     num_files: -1
 
     # Embeddings configuration
-    embeddings_save_loc: "embeddings"
     embedding_model_name_or_path: "sentence-transformers/all-MiniLM-L6-v2"
     embedding_batch_size: 128
+    embeddings_save_loc: "embeddings"
+    embedding_pooling_strategy: "mean_pooling"
+    embedding_column: "embeddings"
     write_embeddings_to_disk: true
+    write_to_filename: false
 
     # Clustering configuration
-    clustering_save_loc: "clustering_results"
-    n_clusters: 1000
-    seed: 1234
     max_iter: 100
-    kmeans_with_cos_dist: false
-
-    # Semdedup configuration
-    which_to_keep: "hard"
-    largest_cluster_size_to_process: 100000
+    n_clusters: 1000
+    clustering_save_loc: "clustering_results"
     sim_metric: "cosine"
+    which_to_keep: "hard"
+    sort_clusters: true
+    kmeans_with_cos_dist: false
+    partition_size: "2gb"
 
     # Extract dedup configuration
     eps_thresholds:

nemo_curator/modules/config.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2025, NVIDIA CORPORATION.  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.
@@ -147,51 +147,71 @@ class SemDedupConfig(BaseConfig):
 
     Attributes:
         cache_dir (str): Directory to store cache.
-        profile_dir (Optional[str]): If specified directory to write dask profile. Default is None.
-        cache_dir (str): Directory to store cache.
+        profile_dir (Optional[str]): If specified, directory to write Dask profile.
+            Default is None.
         num_files (int): Number of files. Default is -1, meaning all files.
-        embeddings_save_loc (str): Location to save embeddings.
+
         embedding_model_name_or_path (str): Model name or path for embeddings.
-        embedding_batch_size (int): Inital Batch size for processing embeddings.
-        embedding_pooling_strategy (str): Strategy for pooling embeddings, either "mean_pooling" or "last_token". Defaults to "mean_pooling".
-        write_embeddings_to_disk (bool): If True, saves the embeddings to disk, defaults to True.
+            Default is "sentence-transformers/all-MiniLM-L6-v2".
+        embedding_batch_size (int): Initial batch size for processing embeddings.
+            Default is 128.
+        embeddings_save_loc (str): Location to save embeddings.
+            Default is "embeddings".
+        embedding_max_mem_gb (int): Maximum memory usage in GB for the embedding process.
+            If None, it defaults to the available GPU memory minus 4 GB.
+        embedding_pooling_strategy (str): Strategy for pooling embeddings, either
+            "mean_pooling" or "last_token". Default is "mean_pooling".
+        embedding_column (str): The column name that stores the embeddings.
+            Default is "embeddings".
+        write_embeddings_to_disk (bool): If True, saves the embeddings to disk.
             We recommend setting this to False when you have a delayed pipeline.
-            Setting it to False can lead to more memory overhead.
+            Setting it to False can lead to more memory overhead. Default is True.
+        write_to_filename (bool): If True, saves the embeddings to the same filename as input files.
+            Default False.
+
+        max_iter (int): Maximum iterations for clustering. Default is 100.
+        n_clusters (int): Number of clusters. Default is 1000.
         clustering_save_loc (str): Location to save clustering results.
-        n_clusters (int): Number of clusters.
-        seed (int): Seed for clustering.
-        max_iter (int): Maximum iterations for clustering.
-        kmeans_with_cos_dist (bool): Use KMeans with cosine distance.
-        which_to_keep (str): Which duplicates to keep.
-        largest_cluster_size_to_process (int): Largest cluster size to process.
+            Default is "clustering_results".
         sim_metric (str): Similarity metric for deduplication.
-        eps_thresholds (List[float]): Epsilon thresholds to calculate if semantically similar or not.
+            Default is "cosine".
+        which_to_keep (str): Method to determine which duplicates to keep.
+            Default is "hard".
+        sort_clusters (bool): Whether to sort clusters. Default is True.
+        kmeans_with_cos_dist (bool): Whether or not to use KMeans with cosine distance.
+            Default is False.
+        partition_size (str): The size of data partition with which to run KMeans.
+            Default is "2gb".
+
+        eps_thresholds (List[float]): Epsilon thresholds to calculate if semantically
+            similar or not. Default is [0.01, 0.001].
         eps_to_extract (float): Epsilon value to extract deduplicated data.
+            Default is 0.01.
     """
 
     cache_dir: str
     profile_dir: Optional[str] = None
     num_files: int = -1
 
     # Embeddings
-    embeddings_save_loc: str = "embeddings"
     embedding_model_name_or_path: str = "sentence-transformers/all-MiniLM-L6-v2"
     embedding_batch_size: int = 128
-    # Options: "mean_pooling", "last_token"
+    embeddings_save_loc: str = "embeddings"
+    embedding_max_mem_gb: Optional[int] = None
     embedding_pooling_strategy: str = "mean_pooling"
+    embedding_column: str = "embeddings"
     write_embeddings_to_disk: bool = True
+    write_to_filename: bool = False
 
-    # Clustering config
-    clustering_save_loc: str = "clustering_results"
-    n_clusters: int = 1000
-    seed: int = 1234
+    # Clustering
     max_iter: int = 100
-    kmeans_with_cos_dist: bool = False
-
-    # Semdedup config
-    which_to_keep: str = "hard"
-    largest_cluster_size_to_process: int = 100000
+    n_clusters: int = 1000
+    clustering_save_loc: str = "clustering_results"
     sim_metric: str = "cosine"
+    which_to_keep: str = "hard"
+    sort_clusters: bool = True
+    kmeans_with_cos_dist: bool = False
+    partition_size: str = "2gb"
 
     # Extract dedup config
     eps_thresholds: List[float] = field(default_factory=lambda: [0.01, 0.001])

nemo_curator/modules/semantic_dedup/clusteringmodel.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2025, NVIDIA CORPORATION.  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.
@@ -54,7 +54,7 @@ def __init__(
         max_iter: int,
         n_clusters: int,
         clustering_output_dir: str,
-        embedding_col: str = "embeddings",
+        embedding_column: str = "embeddings",
         sim_metric: str = "cosine",
         which_to_keep: str = "hard",
         sort_clusters: bool = True,
@@ -71,7 +71,7 @@ def __init__(
             max_iter (int): Maximum number of iterations for the clustering algorithm.
             n_clusters (int): The number of clusters to form.
             clustering_output_dir (str): Directory path where clustering results will be saved.
-            embedding_col (str): Column name where the embeddings are stored.
+            embedding_column (str): Column name where the embeddings are stored.
             sim_metric (str): Similarity metric to use for clustering, default is "cosine".
             which_to_keep (str): Strategy to decide which duplicates to keep; default is "hard".
             sort_clusters (bool): Whether to sort clusters, default is True.
@@ -86,7 +86,7 @@ def __init__(
         self.max_iter = max_iter
         self.n_clusters = n_clusters
         self.clustering_output_dir = clustering_output_dir
-        self.embedding_col = embedding_col
+        self.embedding_column = embedding_column
         self.sim_metric = sim_metric
         self.keep_hard = which_to_keep == "hard"
         self.kmeans_with_cos_dist = kmeans_with_cos_dist
@@ -117,22 +117,37 @@ def _setup_logger(self, logger):
     def __call__(self, embeddings_dataset: DocumentDataset):
         embeddings_df = embeddings_dataset.df
 
-        if self.embedding_col not in embeddings_df.columns:
+        if self.embedding_column not in embeddings_df.columns:
             raise ValueError(
-                f"Expected embedding column '{self.embedding_col}'"
+                f"Expected embedding column '{self.embedding_column}'"
                 f" to be in dataset. Only found columns {embeddings_df.columns}"
             )
 
         with performance_report_if_with_ts_suffix(self.profile_dir, "clustering-model"):
-            embeddings_df = embeddings_df[[self.id_col, self.embedding_col]]
+            embeddings_df = embeddings_df[[self.id_col, self.embedding_column]]
             embeddings_df = embeddings_df.repartition(
                 partition_size=self.partition_size
             )
-            embeddings_df = embeddings_df.to_backend("pandas").persist()
+
+            try:
+                embeddings_df = embeddings_df.to_backend("pandas").persist()
+
+                if embeddings_df.shape[0].compute() < self.n_clusters:
+                    raise ValueError(
+                        "Number of clusters is greater than the number of documents in your dataset. "
+                        "Please reduce n_clusters to be less than or equal."
+                    )
+            except IndexError as e:
+                raise IndexError(
+                    f'Original error message: "{e}". '
+                    "This could be due to empty partitions in your DocumentDataset. "
+                    "Please check your dataset for empty partitions and remove them if necessary."
+                )
+
             embeddings_df = embeddings_df.to_backend("cudf")
 
             cupy_darr = embeddings_df.map_partitions(
-                get_embedding_ar, self.embedding_col, meta=cp.ndarray([1, 1])
+                get_embedding_ar, self.embedding_column, meta=cp.ndarray([1, 1])
             )
             cupy_darr.compute_chunk_sizes()
             t0 = time.time()
@@ -156,7 +171,7 @@ def __call__(self, embeddings_dataset: DocumentDataset):
             meta_df["dist_to_cent"] = cp.zeros(1)
             embeddings_df = embeddings_df.map_partitions(
                 add_dist_to_cents,
-                embedding_col=self.embedding_col,
+                embedding_col=self.embedding_column,
                 centroids=kmeans.cluster_centers_,
                 meta=meta_df,
             )
@@ -198,7 +213,7 @@ def __call__(self, embeddings_dataset: DocumentDataset):
                 output_sorted_clusters_dir=os.path.join(
                     self.clustering_output_dir, "sorted"
                 ),
-                embedding_col=self.embedding_col,
+                embedding_col=self.embedding_column,
                 sim_metric=self.sim_metric,
                 keep_hard=self.keep_hard,
                 kmeans_with_cos_dist=self.kmeans_with_cos_dist,

nemo_curator/modules/semantic_dedup/embeddings.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2025, NVIDIA CORPORATION.  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.
@@ -228,6 +228,12 @@ def __call__(self, dataset: DocumentDataset) -> DocumentDataset:
                 self.profile_dir, "embedding-creator"
             ):
                 embedding_ddf = self.create_embeddings(dataset.df, self.input_column)
+
+                # category column dtypes are not supported by the GPU-accelerated Parquet writer
+                for col in embedding_ddf.columns:
+                    if embedding_ddf[col].dtype.name == "category":
+                        embedding_ddf[col] = embedding_ddf[col].astype("str")
+
                 write_to_disk(
                     embedding_ddf,
                     self.embedding_output_dir,

nemo_curator/modules/semantic_dedup/semanticclusterleveldedup.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2025, NVIDIA CORPORATION.  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.
@@ -42,7 +42,7 @@ def __init__(
         id_column_type: str,
         which_to_keep: str,
         output_dir: str,
-        embedding_col: str = "embeddings",
+        embedding_column: str = "embeddings",
         logger: Union[logging.Logger, str] = "./",
         profile_dir: Optional[str] = None,
     ) -> None:
@@ -57,7 +57,7 @@ def __init__(
             id_column_type (str): Data type of the ID column.
             which_to_keep (str): Strategy for which duplicate to keep.
             output_dir (str): Directory to save output files.
-            embedding_col (str): Column where the embeddings are stored.
+            embedding_column (str): Column where the embeddings are stored.
             logger (Union[logging.Logger, str]): Logger instance or path to the log file directory.
             profile_dir (str): If specified directory to write dask profile. Default is None.
         """
@@ -72,7 +72,7 @@ def __init__(
             output_dir, "semdedup_pruning_tables"
         )
         self.computed_semantic_match_dfs = False
-        self.embedding_col = embedding_col
+        self.embedding_column = embedding_column
         self.logger = self._setup_logger(logger)
         self.profile_dir = profile_dir
 
@@ -132,7 +132,7 @@ def compute_semantic_match_dfs(
                     id_col_type=self.id_col_type,
                     eps_list=eps_list,
                     output_dir=self.semdedup_pruning_tables_dir,
-                    embedding_col=self.embedding_col,
+                    embedding_col=self.embedding_column,
                     which_to_keep=self.which_to_keep,
                 )
             )

nemo_curator/modules/semantic_dedup/semdedup.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2025, NVIDIA CORPORATION.  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.
@@ -50,10 +50,13 @@ def __init__(
         self.embedding_creator = EmbeddingCreator(
             embedding_model_name_or_path=config.embedding_model_name_or_path,
             embedding_batch_size=config.embedding_batch_size,
+            embedding_output_dir=os.path.join(cache_dir, config.embeddings_save_loc),
+            embedding_max_mem_gb=config.embedding_max_mem_gb,
             embedding_pooling_strategy=config.embedding_pooling_strategy,
             input_column=input_column,
-            embedding_output_dir=os.path.join(cache_dir, config.embeddings_save_loc),
+            embedding_column=config.embedding_column,
             write_embeddings_to_disk=config.write_embeddings_to_disk,
+            write_to_filename=config.write_to_filename,
             logger=logger,
             profile_dir=self.config.profile_dir,
         )
@@ -62,6 +65,12 @@ def __init__(
             max_iter=config.max_iter,
             n_clusters=config.n_clusters,
             clustering_output_dir=os.path.join(cache_dir, config.clustering_save_loc),
+            embedding_column=config.embedding_column,
+            sim_metric=config.sim_metric,
+            which_to_keep=config.which_to_keep,
+            sort_clusters=config.sort_clusters,
+            kmeans_with_cos_dist=config.kmeans_with_cos_dist,
+            partition_size=config.partition_size,
             logger=logger,
             profile_dir=self.config.profile_dir,
         )
@@ -77,6 +86,7 @@ def __init__(
             id_column_type=id_column_type,
             which_to_keep=config.which_to_keep,
             output_dir=os.path.join(cache_dir, config.clustering_save_loc),
+            embedding_column=config.embedding_column,
             logger=logger,
             profile_dir=self.config.profile_dir,
         )

nemo_curator/scripts/semdedup/clustering.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.  All rights reserved.
+# Copyright (c) 2025, NVIDIA CORPORATION.  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.
@@ -93,7 +93,6 @@ def attach_args():
             " cache_dir for the directory to store cache,"
             " clustering_save_loc for the location to save clustering results,"
             " n_clusters for the number of clusters,"
-            " seed for the seed for clustering,"
             " max_iter for the maximum iterations for clustering,"
             " kmeans_with_cos_dist for using K-Means with cosine distance."
         ),

nemo_curator/scripts/semdedup/extract_dedup_data.py

@@ -72,7 +72,6 @@ def attach_args():
             "Important configuration parameters include:"
             " cache_dir for the directory to store cache"
             " which_to_keep for specifying which duplicates to keep,"
-            " largest_cluster_size_to_process for the largest cluster size to process,"
             " sim_metric for the similarity metric for deduplication,"
             " eps_thresholds for epsilon thresholds to calculate if semantically similar or not"
             " and eps_to_extract for the epsilon value to extract deduplicated data."