nv-ingest-client 2025.11.17.dev20251117__py3-none-any.whl → 2025.12.17.dev20251217__py3-none-any.whl

nv_ingest_client/util/vdb/lancedb.py

@@ -0,0 +1,276 @@
+import logging
+
+
+from nv_ingest_client.util.vdb.adt_vdb import VDB
+from datetime import timedelta
+from functools import partial
+from urllib.parse import urlparse
+from nv_ingest_client.util.transport import infer_microservice
+import lancedb
+import pyarrow as pa
+
+logger = logging.getLogger(__name__)
+
+
+def create_lancedb_results(results):
+    """Transform NV-Ingest pipeline results into LanceDB ingestible rows.
+
+    The NV-Ingest pipeline provides nested lists of record dictionaries. This
+    helper extracts the inner `metadata` dict for each record, filters out
+    entries without an embedding, and returns a list of dictionaries with the
+    exact fields expected by the LanceDB table schema used in
+    `LanceDB.create_index`.
+
+    Parameters
+    ----------
+    results : list
+        Nested list-of-lists containing record dicts in the NV-Ingest format.
+
+    Returns
+    -------
+    list
+        List of dictionaries with keys: `vector` (embedding list), `text`
+        (string content), `metadata` (page number) and `source` (source id).
+
+    Notes
+    -----
+    - The function expects each inner record to have a `metadata` mapping
+        containing `embedding`, `content`, `content_metadata.page_number`, and
+        `source_metadata.source_id`.
+    - Records with `embedding is None` are skipped.
+    """
+    old_results = [res["metadata"] for result in results for res in result]
+    results = []
+    for result in old_results:
+        if result["embedding"] is None:
+            continue
+        results.append(
+            {
+                "vector": result["embedding"],
+                "text": result["content"],
+                "metadata": result["content_metadata"]["page_number"],
+                "source": result["source_metadata"]["source_id"],
+            }
+        )
+    return results
+
+
+class LanceDB(VDB):
+    """LanceDB operator implementing the VDB interface.
+
+    This class adapts NV-Ingest records to LanceDB, providing index creation,
+    ingestion, and retrieval hooks. The implementation is intentionally small
+    and focuses on the example configuration used in NV-Ingest evaluation
+    scripts.
+    """
+
+    def __init__(
+        self,
+        uri=None,
+        overwrite=True,
+        table_name="nv-ingest",
+        index_type="IVF_HNSW_SQ",
+        metric="l2",
+        num_partitions=16,
+        num_sub_vectors=256,
+        **kwargs
+    ):
+        """Initialize the LanceDB VDB operator.
+
+        Parameters
+        ----------
+        uri: str, optional
+            LanceDB connection URI (default is "lancedb" for local file-based
+            storage).
+        overwrite : bool, optional
+            If True, existing tables will be overwritten during index creation.
+            If False, new data will be appended to existing tables.
+        table_name : str, optional
+            Name of the LanceDB table to create/use (default is "nv-ingest").
+        index_type : str, optional
+            Type of vector index to create (default is "IVF_HNSW_SQ").
+        metric : str, optional
+            Distance metric for the vector index (default is "l2").
+        num_partitions : int, optional
+            Number of partitions for the vector index (default is 16).
+        num_sub_vectors : int, optional
+            Number of sub-vectors for the vector index (default is 256).
+        **kwargs : dict
+            Forwarded configuration options. This implementation does not
+            actively consume specific keys, but passing parameters such as
+            `uri`, `index_name`, or security options is supported by the
+            interface pattern and may be used by future enhancements.
+        """
+        self.uri = uri or "lancedb"
+        self.overwrite = overwrite
+        self.table_name = table_name
+        self.index_type = index_type
+        self.metric = metric
+        self.num_partitions = num_partitions
+        self.num_sub_vectors = num_sub_vectors
+        super().__init__(**kwargs)
+
+    def create_index(self, records=None, table_name="nv-ingest", **kwargs):
+        """Create a LanceDB table and populate it with transformed records.
+
+        This method connects to LanceDB, transforms NV-Ingest records using
+        `create_lancedb_results`, builds a PyArrow schema that matches the
+        expected table layout, and creates/overwrites a table named `bo`.
+
+        Parameters
+        ----------
+        records : list, optional
+            NV-Ingest records in nested list format (the same structure passed
+            to `run`). If ``None``, an empty table will be created.
+
+        table_name : str, optional
+            Name of the LanceDB table to create (default is "nv-ingest").
+
+        Returns
+        -------
+        table
+            The LanceDB table object returned by `db.create_table`.
+        """
+        db = lancedb.connect(uri=self.uri)
+        results = create_lancedb_results(records)
+        schema = pa.schema(
+            [
+                pa.field("vector", pa.list_(pa.float32(), 2048)),
+                pa.field("text", pa.string()),
+                pa.field("metadata", pa.string()),
+                pa.field("source", pa.string()),
+            ]
+        )
+        table = db.create_table(
+            table_name, data=results, schema=schema, mode="overwrite" if self.overwrite else "append"
+        )
+        return table
+
+    def write_to_index(
+        self,
+        records,
+        table=None,
+        index_type="IVF_HNSW_SQ",
+        metric="l2",
+        num_partitions=16,
+        num_sub_vectors=256,
+        **kwargs
+    ):
+        """Create an index on the LanceDB table and wait for it to become ready.
+
+        This function calls `table.create_index` with an IVF+HNSW+SQ index
+        configuration used in NV-Ingest benchmarks. After requesting index
+        construction it lists available indices and waits for each one to
+        reach a ready state using `table.wait_for_index`.
+
+        Parameters
+        ----------
+        records : list
+            The original records being indexed (not used directly in this
+            implementation but kept in the signature for consistency).
+        table : object
+            LanceDB table object returned by `create_index`.
+        """
+        table.create_index(
+            index_type=index_type,
+            metric=metric,
+            num_partitions=num_partitions,
+            num_sub_vectors=num_sub_vectors,
+            # accelerator="cuda",
+            vector_column_name="vector",
+        )
+        for index_stub in table.list_indices():
+            table.wait_for_index([index_stub.name], timeout=timedelta(seconds=600))
+
+    def retrieval(
+        self,
+        queries,
+        table=None,
+        embedding_endpoint="http://localhost:8012/v1",
+        nvidia_api_key=None,
+        model_name="nvidia/llama-3.2-nv-embedqa-1b-v2",
+        result_fields=["text", "metadata", "source"],
+        top_k=10,
+        **kwargs
+    ):
+        """Run similarity search for a list of text queries.
+
+        This method converts textual queries to embeddings by calling the
+        transport helper `infer_microservice` (configured to use an NVIDIA
+        embedding model in the example) and performs a vector search against
+        the LanceDB `table`.
+
+        Parameters
+        ----------
+        queries : list[str]
+            Text queries to be embedded and searched.
+        table : object
+            LanceDB table object with a built vector index.
+        embedding_endpoint : str, optional
+            URL of the embedding microservice (default is
+            "http://localhost:8012/v1").
+        nvidia_api_key : str, optional
+            NVIDIA API key for authentication with the embedding service. If
+            ``None``, no authentication is used.
+        model_name : str, optional
+            Name of the embedding model to use (default is
+            "nvidia/llama-3.2-nv-embedqa-1b-v2").
+        result_fields : list, optional
+            List of field names to retrieve from each hit document (default is
+            `["text", "metadata", "source"]`).
+        top_k : int, optional
+            Number of top results to return per query (default is 10).
+
+        Returns
+        -------
+        list[list[dict]]
+            For each input query, a list of hit documents (each document is a
+            dict with fields such as `text`, `metadata`, and `source`). The
+            example limits each query to 20 results.
+        """
+        embed_model = partial(
+            infer_microservice,
+            model_name=model_name,
+            embedding_endpoint=embedding_endpoint,
+            nvidia_api_key=nvidia_api_key,
+            input_type="query",
+            output_names=["embeddings"],
+            grpc=not ("http" in urlparse(embedding_endpoint).scheme),
+        )
+        results = []
+        query_embeddings = embed_model(queries)
+        for query_embed in query_embeddings:
+            results.append(
+                table.search([query_embed], vector_column_name="vector").select(result_fields).limit(top_k).to_list()
+            )
+        return results
+
+    def run(self, records):
+        """Orchestrate index creation and data ingestion.
+
+        The `run` method is the public entry point used by NV-Ingest pipeline
+        tasks. A minimal implementation first ensures the table exists by
+        calling `create_index` and then kicks off index construction with
+        `write_to_index`.
+
+        Parameters
+        ----------
+        records : list
+            NV-Ingest records to index.
+
+        Returns
+        -------
+        list
+            The original `records` list is returned unchanged to make the
+            operator composable in pipelines.
+        """
+        table = self.create_index(records=records, table_name=self.table_name)
+        self.write_to_index(
+            records,
+            table=table,
+            index_type=self.index_type,
+            metric=self.metric,
+            num_partitions=self.num_partitions,
+            num_sub_vectors=self.num_sub_vectors,
+        )
+        return records

nv_ingest_client/util/vdb/milvus.py

@@ -287,6 +287,10 @@ def create_nvingest_index_params(
     gpu_index: bool = True,
     gpu_search: bool = False,
     local_index: bool = True,
+    intermediate_graph_degree: int = 128,
+    graph_degree: int = 100,
+    m: int = 64,
+    ef_construction: int = 512,
 ) -> IndexParams:
     """
     Creates index params necessary to create an index for a collection. At a minimum,
@@ -326,8 +330,8 @@ def create_nvingest_index_params(
                 index_type="GPU_CAGRA",
                 metric_type="L2",
                 params={
-                    "intermediate_graph_degree": 128,
-                    "graph_degree": 100,
+                    "intermediate_graph_degree": intermediate_graph_degree,
+                    "graph_degree": graph_degree,
                     "build_algo": "NN_DESCENT",
                     "cache_dataset_on_device": "true",
                     "adapt_for_cpu": "false" if gpu_search else "true",
@@ -339,7 +343,7 @@ def create_nvingest_index_params(
                 index_name=DENSE_INDEX_NAME,
                 index_type="HNSW",
                 metric_type="L2",
-                params={"M": 64, "efConstruction": 512},
+                params={"M": m, "efConstruction": ef_construction},
             )
     if sparse and local_index:
         index_params.add_index(
@@ -407,6 +411,10 @@ def create_nvingest_collection(
     recreate_meta: bool = False,
     username: str = None,
     password: str = None,
+    intermediate_graph_degree: int = 128,
+    graph_degree: int = 100,
+    m: int = 64,
+    ef_construction: int = 512,
 ) -> CollectionSchema:
     """
     Creates a milvus collection with an nv-ingest compatible schema under
@@ -457,6 +465,10 @@ def create_nvingest_collection(
         gpu_index=gpu_index,
         gpu_search=gpu_search,
         local_index=local_index,
+        intermediate_graph_degree=intermediate_graph_degree,
+        graph_degree=graph_degree,
+        m=m,
+        ef_construction=ef_construction,
     )
     create_collection(client, collection_name, schema, index_params, recreate=recreate)
     d_idx, s_idx = _get_index_types(index_params, sparse=sparse)
@@ -892,7 +904,7 @@ def stream_insert_milvus(records, client: MilvusClient, collection_name: str, ba
     logger.info(f"streamed {count} records")
 
 
-def wait_for_index(collection_name: str, num_elements: int, client: MilvusClient):
+def wait_for_index(collection_name: str, expected_rows_dict: dict, client: MilvusClient):
     """
     This function waits for the index to be built. It checks
     the indexed_rows of the index and waits for it to be equal
@@ -901,32 +913,28 @@ def wait_for_index(collection_name: str, num_elements: int, client: MilvusClient
     (refer to MilvusClient.refresh_load for bulk inserts).
     """
     client.flush(collection_name)
-    # index_names = utility.list_indexes(collection_name)
     indexed_rows = 0
     # observe dense_index, all indexes get populated simultaneously
-    for index_name in [DENSE_INDEX_NAME]:
-        indexed_rows = 0
-        expected_rows = client.describe_index(collection_name, index_name)["indexed_rows"] + num_elements
-        while indexed_rows < expected_rows:
-            pos_movement = 10  # number of iteration allowed without noticing an increase in indexed_rows
+    for index_name, rows_expected in expected_rows_dict.items():
+        indexed_rows = client.describe_index(collection_name, index_name)["indexed_rows"]
+        while indexed_rows < rows_expected:
+            # 0.5% of rows expected allowed without noticing an increase in indexed_rows
+            pos_movement = start_pos_movement = max((rows_expected - indexed_rows) * 0.005, 10)
             for i in range(20):
-                current_indexed_rows = client.describe_index(collection_name, index_name)["indexed_rows"]
+                prev_indexed_rows = indexed_rows
+                indexed_rows = client.describe_index(collection_name, index_name)["indexed_rows"]
                 time.sleep(1)
-                logger.info(
-                    f"Indexed rows, {collection_name}, {index_name} -  {current_indexed_rows} / {expected_rows}"
-                )
-                if current_indexed_rows == expected_rows:
-                    indexed_rows = current_indexed_rows
+                logger.info(f"Indexed rows, {collection_name}, {index_name} -  {indexed_rows} / {rows_expected}")
+                if indexed_rows == rows_expected:
                     break
                 # check if indexed_rows is staying the same, too many times means something is wrong
-                if current_indexed_rows == indexed_rows:
+                if indexed_rows == prev_indexed_rows:
                     pos_movement -= 1
                 else:
-                    pos_movement = 10
+                    pos_movement = start_pos_movement
                 # if pos_movement is 0, raise an error, means the rows are not getting indexed as expected
                 if pos_movement == 0:
                     raise ValueError(f"Rows are not getting indexed as expected for: {index_name} - {collection_name}")
-                indexed_rows = current_indexed_rows
     return indexed_rows
 
 
@@ -953,6 +961,7 @@ def write_to_nvingest_collection(
     stream: bool = False,
     username: str = None,
     password: str = None,
+    no_wait_index: bool = False,
     **kwargs,
 ):
     """
@@ -1046,15 +1055,22 @@ def write_to_nvingest_collection(
     if num_elements < threshold:
         stream = True
     if stream:
+        # most be accessed/saved before adding new records
+        index_names = utility.list_indexes(collection_name)
+        expected_rows = {}
+        for index_name in index_names:
+            expected_rows[index_name] = (
+                int(client.describe_index(collection_name, index_name)["indexed_rows"]) + num_elements
+            )
         stream_insert_milvus(
             cleaned_records,
             client,
             collection_name,
         )
-        if not local_index:
+        if not local_index and not no_wait_index:
             # Make sure all rows are indexed, decided not to wrap in a timeout because we dont
             # know how long this should take, it is num_elements dependent.
-            wait_for_index(collection_name, num_elements, client)
+            wait_for_index(collection_name, expected_rows, client)
     else:
         minio_client = Minio(minio_endpoint, access_key=access_key, secret_key=secret_key, secure=False)
         bucket_name = bucket_name if bucket_name else ClientConfigSchema().minio_bucket_name
@@ -1968,6 +1984,7 @@ class Milvus(VDB):
         threshold: int = 1000,
         username: str = None,
         password: str = None,
+        no_wait_index: bool = False,
         **kwargs,
     ):
         """
@@ -2005,6 +2022,12 @@ class Milvus(VDB):
         """
         kwargs = locals().copy()
         kwargs.pop("self", None)
+        bucket_name = kwargs.get("bucket_name", None)
+        if bucket_name is not None and bucket_name != ClientConfigSchema().minio_bucket_name:
+            raise ValueError(
+                "You must use the environment variable MINIO_BUCKET to specify bucket_name, detected:",
+                f"`bucket_name`: {bucket_name} and MINIO_BUCKET: {ClientConfigSchema().minio_bucket_name}",
+            )
         super().__init__(**kwargs)
 
     def create_index(self, **kwargs):

{nv_ingest_client-2025.11.17.dev20251117.dist-info → nv_ingest_client-2025.12.17.dev20251217.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest-client
-Version: 2025.11.17.dev20251117
+Version: 2025.12.17.dev20251217
 Summary: Python client for the nv-ingest service
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License
@@ -223,6 +223,7 @@ Requires-Dist: pydantic-settings>2.0.0
 Requires-Dist: requests>=2.28.2
 Requires-Dist: setuptools>=78.1.1
 Requires-Dist: tqdm>=4.67.1
+Requires-Dist: lancedb>=0.25.3
 Provides-Extra: milvus
 Requires-Dist: pymilvus==2.5.10; extra == "milvus"
 Requires-Dist: pymilvus[bulk_writer,model]; extra == "milvus"

{nv_ingest_client-2025.11.17.dev20251117.dist-info → nv_ingest_client-2025.12.17.dev20251217.dist-info}/RECORD

@@ -1,31 +1,31 @@
 nv_ingest_client/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest_client/nv_ingest_cli.py,sha256=84fc0-6TUe-0BMasRIiRH4okfjno4AKCaKvUwJEZ45k,14457
+nv_ingest_client/nv_ingest_cli.py,sha256=qeZJZq_ltnNFiytQNwMY3VAL7nBUXW2HnwMzBGaKQJ0,14452
 nv_ingest_client/cli/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest_client/cli/util/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest_client/cli/util/click.py,sha256=YjQU1uF148FU5D3ozC2m1kkfOOJxO1U8U552-T8PjU4,20029
 nv_ingest_client/cli/util/processing.py,sha256=ULGCYQF1RTDQV_b35YM1WQRqIjR2wQRMJWu41DogagE,6259
 nv_ingest_client/cli/util/system.py,sha256=AQLq0DD2Ns8jRanrKu1tmVBKPA9rl-F3-ZsGI6FXLqE,1105
 nv_ingest_client/client/__init__.py,sha256=eEX9l1qmkLH2lAAZU3eP17SCV06ZjjrshHAB_xbboHA,375
-nv_ingest_client/client/client.py,sha256=3uA54D4Y6lSS-Nvz8R8uzkHkoV8vJu8GPQQRPoc-Uxk,77368
+nv_ingest_client/client/client.py,sha256=Mb5V3nQRg_jzr07-jmK5jwgx3_WmzaGmGXrEKfoyjHU,82103
 nv_ingest_client/client/ingest_job_handler.py,sha256=4exvMwXbzwC-tb0dWleXE-AwhJkvxvhkf_u_1bJt30U,18387
-nv_ingest_client/client/interface.py,sha256=Y6JnjaRytlBrhgbU6MJYm2dblLvoYxWEB35TETZDSwk,55022
+nv_ingest_client/client/interface.py,sha256=1gmFQ7bVQDiEweChN_Divv1Y87a4cNkEgH2Shp4tIMw,64915
 nv_ingest_client/client/util/processing.py,sha256=Ky7x7QbLn3BlgYwmrmoIc-o1VwmlmrcP9tn7GVTi0t0,2502
 nv_ingest_client/primitives/__init__.py,sha256=3rbpLCI7Bl0pntGatAxXD_V01y6dcLhHFheI3wqet-I,269
 nv_ingest_client/primitives/jobs/__init__.py,sha256=-yohgHv3LcCtSleHSaxjv1oO7nNcMCjN3ZYoOkIypIk,469
-nv_ingest_client/primitives/jobs/job_spec.py,sha256=TBz5u7KRdQjQvqD0mMzwjTK9Jl3p7yTIknQQs0lfnV8,15909
+nv_ingest_client/primitives/jobs/job_spec.py,sha256=qT8d9zxEO4ODAcwIlyU7yN1HSuQbDkhCXhLA9hNOURc,16831
 nv_ingest_client/primitives/jobs/job_state.py,sha256=CEe_oZr4p_MobauWIyhuNrP8y7AUwxhIGBuO7dN-VOQ,5277
 nv_ingest_client/primitives/tasks/__init__.py,sha256=D8X4XuwCxk4g_sMSpNRL1XsjVE1eACYaUdEjSanSEfU,1130
 nv_ingest_client/primitives/tasks/audio_extraction.py,sha256=KD5VvaRm6PYelfofZq_-83CbOmupgosokZzFERI5wDA,3559
-nv_ingest_client/primitives/tasks/caption.py,sha256=I1nOpfGb1Ts7QsElwfayhw-F_UcYqtesS-HaZzeh4rI,2130
+nv_ingest_client/primitives/tasks/caption.py,sha256=w-xPKN77zruUel0md4OA-x2ciELSLY-8Px1ds76gak0,2498
 nv_ingest_client/primitives/tasks/chart_extraction.py,sha256=s5hsljgSXxQMZHGekpAg6OYJ9k3-DHk5NmFpvtKJ6Zs,1493
 nv_ingest_client/primitives/tasks/dedup.py,sha256=qort6p3t6ZJuK_74sfOOLp3vMT3hkB5DAu3467WenyY,1719
 nv_ingest_client/primitives/tasks/embed.py,sha256=ZLk7txs_0OHSjjxvRTYB5jm9RvvXRFo3i32Mj9d2mfc,7048
-nv_ingest_client/primitives/tasks/extract.py,sha256=ec2aKPU9OMOOw-oalQKAPaNRqgkREQ0ByLkFVqutD6E,9339
+nv_ingest_client/primitives/tasks/extract.py,sha256=jTCOSQG1MG0RoQg4DxPgmYgeHQR7O24hmysygkWYyIY,11270
 nv_ingest_client/primitives/tasks/filter.py,sha256=dr6fWnh94i50MsGbrz9m_oN6DJKWIWsp7sMwm6Mjz8A,2617
 nv_ingest_client/primitives/tasks/infographic_extraction.py,sha256=SyTjZQbdVA3QwM5yVm4fUzE4Gu4zm4tAfNLDZMvySV8,1537
 nv_ingest_client/primitives/tasks/ocr_extraction.py,sha256=w4uNITktOs-FLczL4ZzVdQTP4t_Ha-9PzCJWlXeOEN0,1486
 nv_ingest_client/primitives/tasks/split.py,sha256=8UkB3EialsOTEbsOZLxzmnDIfTJzC6uvjNv21IbgAVA,2332
-nv_ingest_client/primitives/tasks/store.py,sha256=nIOnCH8vw4FLCLVBJYnsS5Unc0QmuO_jEtUp7-E9FU4,4199
+nv_ingest_client/primitives/tasks/store.py,sha256=UeIspL_RDPBbUV3gv8SK3tIoYNun8r4cSSMxXvBSaks,4575
 nv_ingest_client/primitives/tasks/table_extraction.py,sha256=wQIC70ZNFt0DNQ1lxfvyR3Ci8hl5uAymHXTC0p6v0FY,1107
 nv_ingest_client/primitives/tasks/task_base.py,sha256=Mrx6kgePJHolYd3Im6mVISXcVgdulLst2MYG5gPov9I,1687
 nv_ingest_client/primitives/tasks/task_factory.py,sha256=uvGQXjgWmeF015jPWmBhiclzfrUf3_yD2PPeirQBczM,3218
@@ -40,17 +40,18 @@ nv_ingest_client/util/process_json_files.py,sha256=YKR-fGT4kM8zO2p8r5tpo5-vvFywk
 nv_ingest_client/util/processing.py,sha256=bAy8it-OUgGFO3pcy6D3ezpyZ6p2DfmoQUGhx3QmVf8,8989
 nv_ingest_client/util/system.py,sha256=DVIRLlEWkpqftqxazCuPNdaFSjQiHGMYcHzBufJSRUM,2216
 nv_ingest_client/util/transport.py,sha256=Kwi3r-EUD5yOInW2rH7tYm2DXnzP3aU9l95V-BbXO90,1836
-nv_ingest_client/util/util.py,sha256=qwJ4MqF8w4-lws76z8iz1V0Hz_ebDYN8yAKyJPGuHuU,15828
+nv_ingest_client/util/util.py,sha256=zvWgIxIeATrtrS8olo_8-fHQ4aDd83yg2SjNDcHIv4g,16805
 nv_ingest_client/util/zipkin.py,sha256=p2tMtTVAqrZGxmAxWKE42wkx7U5KywiX5munI7rJt_k,4473
 nv_ingest_client/util/file_processing/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-nv_ingest_client/util/file_processing/extract.py,sha256=Hjtem4bJWum1bbUPw7_TG-0Z2-7PsH4bBuqTF7bLn88,4794
+nv_ingest_client/util/file_processing/extract.py,sha256=sJBfyv4N2P0-izN4RyCsnSDKuDNugG_tW8XCqN9Uqck,5574
 nv_ingest_client/util/vdb/__init__.py,sha256=ZmoEzeM9LzwwrVvu_DVUnjRNx-x8ahkNeIrSfSKzbAk,513
-nv_ingest_client/util/vdb/adt_vdb.py,sha256=UubzAMSfyrqqpD-OQErpBs25hC2Mw8zGZ4waenGXPOk,515
-nv_ingest_client/util/vdb/milvus.py,sha256=LHZ4Z6fHk8vQUGQFJ3FZ5iay0Ike6Zur-K9yMiPxe44,80141
+nv_ingest_client/util/vdb/adt_vdb.py,sha256=wT3LJMAy2VQu6daXhc3Pte4Ijs6jN-YP6B9-rnuH_FA,10868
+nv_ingest_client/util/vdb/lancedb.py,sha256=mLykdOFkLC5-SpRvHAvt0do9rhyQDqy_H48D6hEtegw,10037
+nv_ingest_client/util/vdb/milvus.py,sha256=NLlsYU5LdESh0r_Psvn0vzGiNN-70iouOGr3RgZaMVg,81316
 nv_ingest_client/util/vdb/opensearch.py,sha256=I4FzF95VWCOkyzhfm-szdfK1Zd9ugUc8AxxpAdEMWGE,7538
-nv_ingest_client-2025.11.17.dev20251117.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-nv_ingest_client-2025.11.17.dev20251117.dist-info/METADATA,sha256=bgCG3WP30zjURzJ_SZEm3fDbby-NoICZDYfbiA3sSjg,30627
-nv_ingest_client-2025.11.17.dev20251117.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-nv_ingest_client-2025.11.17.dev20251117.dist-info/entry_points.txt,sha256=3uQVZkTZIjO08_bjTV-g0CDF5H1nrP1zWXU9gJOweuI,137
-nv_ingest_client-2025.11.17.dev20251117.dist-info/top_level.txt,sha256=1eMhBFD3SiWmpXnod2LM66C1HrSLSk96ninZi5XX-cE,17
-nv_ingest_client-2025.11.17.dev20251117.dist-info/RECORD,,
+nv_ingest_client-2025.12.17.dev20251217.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+nv_ingest_client-2025.12.17.dev20251217.dist-info/METADATA,sha256=EbEZoUk3-GvCBAB2z0hqZjgMOGasw75hZCWTDk7yxpk,30658
+nv_ingest_client-2025.12.17.dev20251217.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+nv_ingest_client-2025.12.17.dev20251217.dist-info/entry_points.txt,sha256=3uQVZkTZIjO08_bjTV-g0CDF5H1nrP1zWXU9gJOweuI,137
+nv_ingest_client-2025.12.17.dev20251217.dist-info/top_level.txt,sha256=1eMhBFD3SiWmpXnod2LM66C1HrSLSk96ninZi5XX-cE,17
+nv_ingest_client-2025.12.17.dev20251217.dist-info/RECORD,,