nv-ingest 2025.6.2.dev20250602__py3-none-any.whl → 2025.7.7.dev20250707__py3-none-any.whl

nv_ingest/framework/orchestration/ray/examples/pipeline_test_harness.py

@@ -63,7 +63,7 @@ def get_nim_service(env_var_prefix):
         "",
     )
     auth_token = os.environ.get(
-        "NVIDIA_BUILD_API_KEY",
+        "NVIDIA_API_KEY",
         "",
     ) or os.environ.get(
         "NGC_API_KEY",
@@ -151,11 +151,11 @@ if __name__ == "__main__":
     os.environ["PADDLE_INFER_PROTOCOL"] = "grpc"
     os.environ["NEMORETRIEVER_PARSE_HTTP_ENDPOINT"] = "https://integrate.api.nvidia.com/v1/chat/completions"
     os.environ["VLM_CAPTION_ENDPOINT"] = "https://integrate.api.nvidia.com/v1/chat/completions"
-    os.environ["VLM_CAPTION_MODEL_NAME"] = "meta/llama-3.2-11b-vision-instruct"
+    os.environ["VLM_CAPTION_MODEL_NAME"] = "nvidia/llama-3.1-nemotron-nano-vl-8b-v1"
     logger.info("Environment variables set.")
 
     image_caption_endpoint_url = "https://integrate.api.nvidia.com/v1/chat/completions"
-    image_caption_model_name = "meta/llama-3.2-11b-vision-instruct"
+    model_name = "nvidia/llama-3.1-nemotron-nano-vl-8b-v1"
     yolox_grpc, yolox_http, yolox_auth, yolox_protocol = get_nim_service("yolox")
     (
         yolox_table_structure_grpc,
@@ -228,7 +228,7 @@ if __name__ == "__main__":
     image_caption_config = {
         "api_key": yolox_auth,
         "endpoint_url": image_caption_endpoint_url,
-        "image_caption_model_name": image_caption_model_name,
+        "model_name": model_name,
         "prompt": "Caption the content of this image:",
     }
     logger.info("Service configuration retrieved from get_nim_service and environment variables.")

nv_ingest/framework/orchestration/ray/primitives/pipeline_topology.py

@@ -555,7 +555,7 @@ class PipelineTopology:
             return None
 
     def get_connections(self) -> Dict[str, List[Tuple[str, int]]]:
-        """Returns a shallow copy of the connections dictionary."""
+        """Returns a shallow copy of the connection dictionary."""
         with self._lock:
             # Shallow copy is usually sufficient here as tuples are immutable
             return self._connections.copy()
@@ -571,7 +571,7 @@ class PipelineTopology:
             return len(self._stage_actors.get(stage_name, []))
 
     def get_edge_queues(self) -> Dict[str, Tuple[Any, int]]:
-        """Returns a shallow copy of the edge queues dictionary."""
+        """Returns a shallow copy of the edge queues' dictionary."""
         with self._lock:
             return self._edge_queues.copy()
 

nv_ingest/framework/orchestration/ray/primitives/ray_pipeline.py

@@ -9,6 +9,7 @@ import threading
 from abc import ABC, abstractmethod
 from collections import defaultdict
 from dataclasses import dataclass
+from types import FunctionType
 
 import psutil
 import uuid
@@ -24,6 +25,9 @@ import time
 from nv_ingest.framework.orchestration.ray.primitives.pipeline_topology import PipelineTopology, StageInfo
 from nv_ingest.framework.orchestration.ray.primitives.ray_stat_collector import RayStatsCollector
 from nv_ingest.framework.orchestration.ray.util.pipeline.pid_controller import PIDController, ResourceConstraintManager
+from nv_ingest.framework.orchestration.ray.util.pipeline.tools import wrap_callable_as_stage
+from nv_ingest_api.util.imports.callable_signatures import ingest_stage_callable_signature
+from nv_ingest_api.util.imports.dynamic_resolvers import resolve_callable_from_path
 
 logger = logging.getLogger(__name__)
 
@@ -43,7 +47,7 @@ class PipelineInterface(ABC):
         Parameters
         ----------
         monitor_poll_interval : float
-            Interval in seconds for monitoring poll (default: 5.0).
+            Interval in seconds for the monitoring poll (default: 5.0).
         scaling_poll_interval : float
             Interval in seconds for scaling decisions (default: 30.0).
         """
@@ -270,7 +274,7 @@ class RayPipeline(PipelineInterface):
 
         logger.info("RayStatsCollector initialized using StatsConfig.")
 
-    # --- Accessor Methods for Stats Collector (and internal use) ---
+    # --- Accessor Methods for Stat Collector (and internal use) ---
 
     def __del__(self):
         try:
@@ -428,15 +432,39 @@ class RayPipeline(PipelineInterface):
         return self
 
     def add_stage(
-        self, *, name: str, stage_actor: Any, config: BaseModel, min_replicas: int = 0, max_replicas: int = 1
+        self,
+        *,
+        name: str,
+        stage_actor: Any,
+        config: BaseModel,
+        min_replicas: int = 0,
+        max_replicas: int = 1,
     ) -> "RayPipeline":
         if min_replicas < 0:
             logger.warning(f"Stage '{name}': min_replicas cannot be negative. Overriding to 0.")
             min_replicas = 0
+
+        resolved_actor = stage_actor
+
+        # Support module path (e.g., "mypkg.mymodule:my_lambda")
+        if isinstance(stage_actor, str):
+            resolved_actor = resolve_callable_from_path(
+                callable_path=stage_actor, signature_schema=ingest_stage_callable_signature
+            )
+
+        # Wrap callables
+        if isinstance(resolved_actor, FunctionType):
+            schema_type = type(config)
+            resolved_actor = wrap_callable_as_stage(resolved_actor, schema_type)
+
         stage_info = StageInfo(
-            name=name, callable=stage_actor, config=config, min_replicas=min_replicas, max_replicas=max_replicas
+            name=name,
+            callable=resolved_actor,
+            config=config,
+            min_replicas=min_replicas,
+            max_replicas=max_replicas,
         )
-        self.topology.add_stage(stage_info)  # Delegate
+        self.topology.add_stage(stage_info)
 
         return self
 

nv_ingest/framework/orchestration/ray/primitives/ray_stat_collector.py

@@ -40,7 +40,7 @@ class RayStatsCollector:
             - `get_edge_queues() -> Dict[str, Tuple[Any, int]]`
             These methods should return snapshots suitable for iteration.
         interval : float, optional
-            The interval in seconds between stats collection attempts, by default 5.0.
+            The interval in seconds between stat collection attempts, by default 5.0.
         actor_timeout : float, optional
             Timeout in seconds for waiting for stats from a single actor, by default 5.0.
         queue_timeout : float, optional

nv_ingest/framework/orchestration/ray/stages/injectors/metadata_injector.py

@@ -2,6 +2,7 @@
 # All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+from datetime import datetime
 import logging
 import pandas as pd
 from typing import Any
@@ -9,8 +10,15 @@ from pydantic import BaseModel
 import ray
 
 from nv_ingest.framework.orchestration.ray.stages.meta.ray_actor_stage_base import RayActorStage
-from nv_ingest_api.internal.enums.common import DocumentTypeEnum, ContentTypeEnum
+from nv_ingest_api.internal.enums.common import (
+    DocumentTypeEnum,
+    ContentTypeEnum,
+    AccessLevelEnum,
+    TextTypeEnum,
+    LanguageEnum,
+)
 from nv_ingest_api.internal.primitives.tracing.tagging import traceable
+from nv_ingest_api.internal.schemas.meta.metadata_schema import ContentHierarchySchema
 from nv_ingest_api.util.converters.type_mappings import doc_type_to_content_type
 from nv_ingest_api.util.exception_handlers.decorators import (
     nv_ingest_node_failure_try_except,
@@ -61,27 +69,83 @@ class MetadataInjectionStage(RayActorStage):
                 # Convert document type to content type using enums.
                 content_type = doc_type_to_content_type(DocumentTypeEnum(row["document_type"]))
                 # Check if metadata is missing or doesn't contain 'content'
-                if "metadata" not in row or not isinstance(row["metadata"], dict) or "content" not in row["metadata"]:
+                if (
+                    "metadata" not in row
+                    or not isinstance(row["metadata"], dict)
+                    or "content" not in row["metadata"].keys()
+                ):
                     update_required = True
+
+                    # Initialize default structures based on MetaDataSchema
+                    default_source_metadata = {
+                        "source_id": row.get("source_id"),
+                        "source_name": row.get("source_name"),
+                        "source_type": row["document_type"],
+                        "source_location": "",
+                        "collection_id": "",
+                        "date_created": datetime.now().isoformat(),
+                        "last_modified": datetime.now().isoformat(),
+                        "summary": "",
+                        "partition_id": -1,
+                        "access_level": AccessLevelEnum.UNKNOWN.value,
+                    }
+
+                    default_content_metadata = {
+                        "type": content_type.name.lower(),
+                        "page_number": -1,
+                        "description": "",
+                        "hierarchy": ContentHierarchySchema().model_dump(),
+                        "subtype": "",
+                        "start_time": -1,
+                        "end_time": -1,
+                    }
+
+                    default_audio_metadata = None
+                    if content_type == ContentTypeEnum.AUDIO:
+                        default_audio_metadata = {
+                            "audio_type": row["document_type"],
+                            "audio_transcript": "",
+                        }
+
+                    default_image_metadata = None
+                    if content_type == ContentTypeEnum.IMAGE:
+                        default_image_metadata = {
+                            "image_type": row["document_type"],
+                            "structured_image_type": ContentTypeEnum.NONE.value,
+                            "caption": "",
+                            "text": "",
+                            "image_location": (0, 0, 0, 0),
+                            "image_location_max_dimensions": (0, 0),
+                            "uploaded_image_url": "",
+                            "width": 0,
+                            "height": 0,
+                        }
+
+                    default_text_metadata = None
+                    if content_type == ContentTypeEnum.TEXT:
+                        default_text_metadata = {
+                            "text_type": TextTypeEnum.DOCUMENT.value,
+                            "summary": "",
+                            "keywords": "",
+                            "language": LanguageEnum.UNKNOWN.value,
+                            "text_location": (0, 0, 0, 0),
+                            "text_location_max_dimensions": (0, 0, 0, 0),
+                        }
+
                     row["metadata"] = {
-                        "content": row.get("content"),
-                        "content_metadata": {
-                            "type": content_type.name.lower(),
-                        },
+                        "content": row["content"],
+                        "content_metadata": default_content_metadata,
                         "error_metadata": None,
-                        "audio_metadata": (
-                            None if content_type != ContentTypeEnum.AUDIO else {"audio_type": row["document_type"]}
-                        ),
-                        "image_metadata": (
-                            None if content_type != ContentTypeEnum.IMAGE else {"image_type": row["document_type"]}
-                        ),
-                        "source_metadata": {
-                            "source_id": row.get("source_id"),
-                            "source_name": row.get("source_name"),
-                            "source_type": row["document_type"],
-                        },
-                        "text_metadata": (None if content_type != ContentTypeEnum.TEXT else {"text_type": "document"}),
+                        "audio_metadata": default_audio_metadata,
+                        "image_metadata": default_image_metadata,
+                        "source_metadata": default_source_metadata,
+                        "text_metadata": default_text_metadata,
                     }
+                    logger.info(
+                        f"METADATA_INJECTOR_DEBUG: Rebuilt metadata for source_id='{row.get('source_id', 'N/A')}'. "
+                        f"Metadata keys: {list(row['metadata'].keys())}."
+                        f"'content' present: {'content' in row['metadata']}"
+                    )
             except Exception as inner_e:
                 logger.exception("Failed to process row during metadata injection")
                 raise inner_e

nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_source_stage_base.py

@@ -23,16 +23,13 @@ class RayActorSourceStage(RayActorStage, ABC):
         super().__init__(config, log_to_stdout=log_to_stdout)
         self.paused = False
 
+    def on_data(self, IngestControlMessage):
+        return NotImplemented("Source stages do not implement on_data().")
+
     @ray.method(num_returns=1)
     def set_input_queue(self, queue_handle: Any) -> bool:
         raise NotImplementedError("Source stages do not support an input queue.")
 
-    def get_input(self) -> Any:
-        """
-        Source stages must implement get_input() to fetch control messages from an external source.
-        """
-        pass
-
     @abstractmethod
     def _read_input(self) -> Any:
         """

nv_ingest/framework/orchestration/ray/stages/sources/message_broker_task_source.py

@@ -304,14 +304,6 @@ class MessageBrokerTaskSourceStage(RayActorSourceStage):
 
         return control_message
 
-    def on_data(self, control_message: any) -> any:
-        """
-        Process the control message.
-        For this source stage, no additional processing is done, so simply return it.
-        """
-        self._logger.debug("on_data: Received control message for processing")
-        return control_message
-
     # In the processing loop, instead of checking a boolean, we wait on the event.
     def _processing_loop(self) -> None:
         """
@@ -336,7 +328,6 @@ class MessageBrokerTaskSourceStage(RayActorSourceStage):
                 self._active_processing = True
 
                 self._logger.debug("Control message received; processing data")
-                updated_cm = self.on_data(control_message)
 
                 # Block until not paused using the pause event.
                 if self.output_queue is not None:
@@ -349,7 +340,7 @@ class MessageBrokerTaskSourceStage(RayActorSourceStage):
 
                     while True:
                         try:
-                            self.output_queue.put(updated_cm)
+                            self.output_queue.put(control_message)
                             self.stats["successful_queue_writes"] += 1
                             break
                         except Exception:

nv_ingest/framework/orchestration/ray/stages/transforms/text_embed.py

@@ -32,7 +32,7 @@ class TextEmbeddingTransformStage(RayActorStage):
     """
 
     def __init__(self, config: TextEmbeddingSchema) -> None:
-        super().__init__(config)
+        super().__init__(config, log_to_stdout=False)
         try:
             self.validated_config = config
             logger.info("TextEmbeddingTransformStage configuration validated successfully.")

nv_ingest/framework/orchestration/ray/stages/transforms/text_splitter.py

@@ -1,11 +1,14 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024-25, NVIDIA CORPORATION & AFFILIATES.
+# All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
 import logging
 from typing import Any
 import ray
 
-# Assume these imports come from your project:
 from nv_ingest.framework.orchestration.ray.stages.meta.ray_actor_stage_base import RayActorStage
 from nv_ingest.framework.util.flow_control import filter_by_task
-from nv_ingest_api.internal.primitives.ingest_control_message import remove_task_by_type
+from nv_ingest_api.internal.primitives.ingest_control_message import remove_task_by_type, IngestControlMessage
 from nv_ingest_api.internal.primitives.tracing.tagging import traceable
 from nv_ingest_api.internal.schemas.transform.transform_text_splitter_schema import TextSplitterSchema
 from nv_ingest_api.internal.transform.split_text import transform_text_split_and_tokenize_internal
@@ -72,3 +75,45 @@ class TextSplitterStage(RayActorStage):
         logger.info("TextSplitterStage.on_data: Finished processing, returning updated message.")
 
         return message
+
+
+def text_splitter_fn(control_message: IngestControlMessage, stage_config: TextSplitterSchema) -> IngestControlMessage:
+    """
+    Process an incoming IngestControlMessage by splitting and tokenizing its text.
+
+    Parameters
+    ----------
+    control_message : IngestControlMessage
+        The incoming message containing the payload DataFrame.
+
+    stage_config : BaseModel
+        The stage level configuration object
+
+    Returns
+    -------
+    IngestControlMessage
+        The updated message with its payload transformed.
+    """
+
+    # Extract the DataFrame payload.
+    df_payload = control_message.payload()
+    logger.debug("Extracted payload with %d rows.", len(df_payload))
+
+    # Remove the "split" task to obtain task-specific configuration.
+    task_config = remove_task_by_type(control_message, "split")
+    logger.debug("Extracted task config: %s", task_config)
+
+    # Transform the DataFrame (split text and tokenize).
+    df_updated = transform_text_split_and_tokenize_internal(
+        df_transform_ledger=df_payload,
+        task_config=task_config,
+        transform_config=stage_config,
+        execution_trace_log=None,
+    )
+    logger.info("TextSplitterStage.on_data: Transformation complete. Updated payload has %d rows.", len(df_updated))
+
+    # Update the message payload.
+    control_message.payload(df_updated)
+    logger.info("TextSplitterStage.on_data: Finished processing, returning updated message.")
+
+    return control_message

nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_builders.py

@@ -174,9 +174,9 @@ def setup_ingestion_pipeline(pipeline: RayPipeline, ingest_config: Dict[str, Any
     pipeline.make_edge(image_dedup_stage_id, text_splitter_stage_id, queue_size=ingest_edge_buffer_size)
 
     ###### Primitive Transforms ########
-    pipeline.make_edge(text_splitter_stage_id, embed_extractions_stage_id, queue_size=ingest_edge_buffer_size)
-    pipeline.make_edge(embed_extractions_stage_id, image_caption_stage_id, queue_size=ingest_edge_buffer_size)
-    pipeline.make_edge(image_caption_stage_id, image_storage_stage_id, queue_size=ingest_edge_buffer_size)
+    pipeline.make_edge(text_splitter_stage_id, image_caption_stage_id, queue_size=ingest_edge_buffer_size)
+    pipeline.make_edge(image_caption_stage_id, embed_extractions_stage_id, queue_size=ingest_edge_buffer_size)
+    pipeline.make_edge(embed_extractions_stage_id, image_storage_stage_id, queue_size=ingest_edge_buffer_size)
 
     ###### Primitive Storage ########
     pipeline.make_edge(image_storage_stage_id, embedding_storage_stage_id, queue_size=ingest_edge_buffer_size)

nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_runners.py

@@ -72,7 +72,7 @@ class PipelineCreationSchema(BaseModel):
 
     # API keys
     ngc_api_key: str = os.getenv("NGC_API_KEY", "")
-    nvidia_build_api_key: str = os.getenv("NVIDIA_BUILD_API_KEY", "")
+    nvidia_api_key: str = os.getenv("NVIDIA_API_KEY", "")
 
     # Observability settings
     otel_exporter_otlp_endpoint: str = os.getenv("OTEL_EXPORTER_OTLP_ENDPOINT", "localhost:4317")
@@ -86,9 +86,9 @@ class PipelineCreationSchema(BaseModel):
 
     # Vision language model settings
     vlm_caption_endpoint: str = os.getenv(
-        "VLM_CAPTION_ENDPOINT", "https://ai.api.nvidia.com/v1/gr/meta/llama-3.2-11b-vision-instruct/chat/completions"
+        "VLM_CAPTION_ENDPOINT", "https://ai.api.nvidia.com/v1/gr/nvidia/llama-3.1-nemotron-nano-vl-8b-v1/chat/completions"
     )
-    vlm_caption_model_name: str = os.getenv("VLM_CAPTION_MODEL_NAME", "meta/llama-3.2-11b-vision-instruct")
+    vlm_caption_model_name: str = os.getenv("VLM_CAPTION_MODEL_NAME", "nvidia/llama-3.1-nemotron-nano-vl-8b-v1")
 
     # YOLOX image processing settings
     yolox_graphic_elements_http_endpoint: str = os.getenv(
@@ -331,6 +331,10 @@ def run_pipeline(
     """
     if run_in_subprocess:
         logger.info("Launching pipeline in Python subprocess using multiprocessing.")
+        if (ingest_config.ngc_api_key is None or ingest_config.ngc_api_key == "") and (
+            ingest_config.nvidia_api_key is None or ingest_config.nvidia_api_key == ""
+        ):
+            logger.warning("NGC_API_KEY or NVIDIA_API_KEY are not set. NIM Related functions will not work.")
 
         ctx = multiprocessing.get_context("fork")
         process = ctx.Process(

nv_ingest/framework/orchestration/ray/util/pipeline/stage_builders.py

@@ -2,8 +2,6 @@
 # All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
-# TODO(Devin)
-# flake8: noqa
 import os
 
 import click
@@ -11,6 +9,7 @@ import logging
 
 from nv_ingest.framework.orchestration.ray.stages.sinks.default_drain import DefaultDrainSink
 from nv_ingest.framework.orchestration.ray.stages.telemetry.otel_tracer import OpenTelemetryTracerStage
+from nv_ingest.framework.orchestration.ray.stages.transforms.text_splitter import TextSplitterStage
 from nv_ingest.framework.schemas.framework_otel_tracer_schema import OpenTelemetryTracerSchema
 from nv_ingest_api.internal.schemas.extract.extract_infographic_schema import InfographicExtractorSchema
 
@@ -41,7 +40,6 @@ from nv_ingest.framework.orchestration.ray.stages.storage.image_storage import I
 from nv_ingest.framework.orchestration.ray.stages.storage.store_embeddings import EmbeddingStorageStage
 from nv_ingest.framework.orchestration.ray.stages.transforms.image_caption import ImageCaptionTransformStage
 from nv_ingest.framework.orchestration.ray.stages.transforms.text_embed import TextEmbeddingTransformStage
-from nv_ingest.framework.orchestration.ray.stages.transforms.text_splitter import TextSplitterStage
 from nv_ingest.framework.schemas.framework_metadata_injector_schema import MetadataInjectorSchema
 from nv_ingest_api.internal.schemas.extract.extract_audio_schema import AudioExtractorSchema
 from nv_ingest_api.internal.schemas.extract.extract_chart_schema import ChartExtractorSchema
@@ -107,7 +105,7 @@ def get_nim_service(env_var_prefix):
         "",
     )
     auth_token = os.environ.get(
-        "NVIDIA_BUILD_API_KEY",
+        "NVIDIA_API_KEY",
         "",
     ) or os.environ.get(
         "NGC_API_KEY",
@@ -137,7 +135,7 @@ def get_audio_retrieval_service(env_var_prefix):
         "",
     )
     auth_token = os.environ.get(
-        "NVIDIA_BUILD_API_KEY",
+        "NVIDIA_API_KEY",
         "",
     ) or os.environ.get(
         "NGC_API_KEY",
@@ -465,7 +463,7 @@ def add_text_splitter_stage(pipeline, default_cpu_count, stage_name="text_splitt
 
 def add_image_caption_stage(pipeline, default_cpu_count, stage_name="image_caption"):
     auth_token = os.environ.get(
-        "NVIDIA_BUILD_API_KEY",
+        "NVIDIA_API_KEY",
         "",
     ) or os.environ.get(
         "NGC_API_KEY",
@@ -473,13 +471,13 @@ def add_image_caption_stage(pipeline, default_cpu_count, stage_name="image_capti
     )
 
     endpoint_url = os.environ.get("VLM_CAPTION_ENDPOINT", "localhost:5000")
-    model_name = os.environ.get("VLM_CAPTION_MODEL_NAME", "meta/llama-3.2-11b-vision-instruct")
+    model_name = os.environ.get("VLM_CAPTION_MODEL_NAME", "nvidia/llama-3.1-nemotron-nano-vl-8b-v1")
 
     config = ImageCaptionExtractionSchema(
         **{
             "api_key": auth_token,
             "endpoint_url": endpoint_url,
-            "image_caption_model_name": model_name,
+            "model_name": model_name,
             "prompt": "Caption the content of this image:",
         }
     )
@@ -497,7 +495,7 @@ def add_image_caption_stage(pipeline, default_cpu_count, stage_name="image_capti
 
 def add_text_embedding_stage(pipeline, default_cpu_count, stage_name="text_embedding"):
     api_key = os.environ.get(
-        "NVIDIA_BUILD_API_KEY",
+        "NVIDIA_API_KEY",
         "",
     ) or os.environ.get(
         "NGC_API_KEY",

nv_ingest/framework/orchestration/ray/util/pipeline/tools.py

@@ -0,0 +1,203 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024-25, NVIDIA CORPORATION & AFFILIATES.
+# All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+import logging
+import uuid
+from typing import Callable, Optional, Union, Dict, List, Type
+
+import ray
+from pydantic import BaseModel
+
+from nv_ingest.framework.orchestration.ray.stages.meta.ray_actor_stage_base import RayActorStage
+from nv_ingest.framework.util.flow_control import filter_by_task
+from nv_ingest_api.internal.primitives.tracing.tagging import traceable
+from nv_ingest_api.util.exception_handlers.decorators import nv_ingest_node_failure_try_except
+
+logger = logging.getLogger(__name__)
+
+
+def wrap_callable_as_stage(
+    fn: Callable[[object, BaseModel], object],
+    schema_type: Type[BaseModel],
+    *,
+    required_tasks: Optional[List[str]] = None,
+    trace_id: Optional[str] = None,
+):
+    """
+    Factory to wrap a user-supplied function into a Ray actor, returning a proxy
+    for unique, isolated dynamic actor creation.
+
+    Parameters
+    ----------
+    fn : Callable[[IngestControlMessage, BaseModel], IngestControlMessage]
+        The processing function to be wrapped in the Ray actor.
+    schema_type : Type[BaseModel]
+        Pydantic schema used to validate and pass the stage config.
+    required_tasks : Optional[List[str]], optional
+        Task names this stage should filter on. If None, no filtering is applied.
+    trace_id : Optional[str], optional
+        Optional name for trace annotation; defaults to the function name.
+
+    Returns
+    -------
+    StageProxy : object
+        A factory-like proxy exposing `.remote()` and `.options()` for Ray-idiomatic
+        actor creation. Direct instantiation or class method use is not supported.
+
+    Notes
+    -----
+    - Each call to `.remote()` or `.options()` generates a new, dynamically created class
+      (using `type()`), ensuring Ray treats each as a unique actor type and preventing
+      class/actor name collisions or registry issues. This is essential when running
+      dynamic or parallel pipelines and tests.
+    - Only `.remote(config)` and `.options(...)` (chained with `.remote(config)`) are supported.
+      All other class/actor patterns will raise `NotImplementedError`.
+    """
+    trace_name = trace_id or fn.__name__
+
+    def make_actor_class():
+        """
+        Dynamically constructs a unique Ray actor class for every call.
+
+        Engineering Note
+        ----------------
+        This pattern uses Python's `type()` to create a new class object for each actor instance,
+        guaranteeing a unique type each time. Ray's internal registry identifies actor types
+        by their Python class object. If you use the same class (even with different logic or
+        @ray.remote), Ray may reuse or overwrite them, causing hard-to-diagnose bugs in
+        parallel or test code. By generating a fresh class each time, we fully isolate state,
+        serialization, and Ray's registry—avoiding actor collisions and test pollution.
+
+        Returns
+        -------
+        new_class : type
+            The dynamically constructed RayActorStage subclass.
+        """
+        class_name = f"LambdaStage_{fn.__name__}_{uuid.uuid4().hex[:8]}"
+
+        def __init__(self, config: Union[Dict, BaseModel]) -> None:
+            """
+            Parameters
+            ----------
+            config : Union[Dict, BaseModel]
+                Stage configuration, validated against `schema_type`.
+            """
+            validated_config = schema_type(**config) if not isinstance(config, schema_type) else config
+            super(new_class, self).__init__(validated_config, log_to_stdout=True)
+            self.validated_config = validated_config
+            self._logger.info(f"{self.__class__.__name__} initialized with validated config.")
+
+        @traceable(trace_name)
+        @nv_ingest_node_failure_try_except(annotation_id=trace_name, raise_on_failure=False)
+        @filter_by_task(required_tasks=required_tasks) if required_tasks else (lambda f: f)
+        def on_data(self, control_message):
+            """
+            Processes a control message using the wrapped function.
+
+            Parameters
+            ----------
+            control_message : IngestControlMessage
+                The message to be processed.
+
+            Returns
+            -------
+            IngestControlMessage
+                The processed message, or the original on failure.
+            """
+            try:
+                return fn(control_message, self.validated_config)
+            except Exception as e:
+                self._logger.exception(f"{self.__class__.__name__} failed: {e}")
+                self.stats["errors"] += 1
+                return control_message
+
+        # --- ENGINEERING NOTE ---
+        # The `class_dict` collects all the methods and attributes for the dynamic class.
+        # This allows us to build a fresh class object per call, preventing Ray from
+        # reusing or overwriting global actor types. It is the critical piece for
+        # robust dynamic actor creation in Ray!
+        # ------------------------
+
+        class_dict = {
+            "__init__": __init__,
+            "on_data": on_data,
+        }
+        bases = (RayActorStage,)
+        new_class = type(class_name, bases, class_dict)
+        return new_class
+
+    class StageProxy:
+        """
+        Factory/proxy for dynamic Ray actor creation; not itself a Ray actor.
+
+        Methods
+        -------
+        remote(config)
+            Instantiate a Ray actor with a unique dynamic class and name.
+        options(*args, **kwargs)
+            Advanced Ray actor configuration (chain with `.remote(config)`).
+        actor_class()
+            Generates and returns a fresh actor class (for introspection/testing only).
+        """
+
+        @staticmethod
+        def remote(config):
+            """
+            Instantiate a Ray actor with a unique dynamic class and name.
+
+            Parameters
+            ----------
+            config : Union[Dict, BaseModel]
+                Stage configuration to pass to the actor.
+
+            Returns
+            -------
+            ray.actor.ActorHandle
+                Handle to the started Ray actor.
+            """
+            _ActorClass = ray.remote(make_actor_class())
+            unique_name = f"{fn.__name__}_{str(uuid.uuid4())[:8]}"
+            return _ActorClass.options(name=unique_name).remote(config)
+
+        @staticmethod
+        def options(*args, **kwargs):
+            """
+            Return a Ray actor class with the specified options set.
+            Must call `.remote(config)` on the result.
+
+            Parameters
+            ----------
+            *args
+                Positional arguments for Ray actor options.
+            **kwargs
+                Keyword arguments for Ray actor options (e.g., resources).
+
+            Returns
+            -------
+            ray.actor.ActorClass
+                Ray actor class, requires .remote(config) to instantiate.
+            """
+            ActorClass = ray.remote(make_actor_class())
+            if "name" not in kwargs:
+                kwargs["name"] = f"{fn.__name__}_{str(uuid.uuid4())[:8]}"
+            return ActorClass.options(*args, **kwargs)
+
+        def __new__(cls, *a, **k):
+            raise NotImplementedError("StageProxy is a factory, not a Ray actor or class. Use .remote() or .options().")
+
+        def __call__(self, *a, **k):
+            raise NotImplementedError("StageProxy is a factory, not a Ray actor or class. Use .remote() or .options().")
+
+        def __getattr__(self, name):
+            # Only allow access to known public members
+            if name in {"remote", "options", "actor_class"}:
+                return getattr(self, name)
+            raise NotImplementedError(
+                f"StageProxy does not implement '{name}'. Only .remote(), .options(), .actor_class() are available."
+            )
+
+        # For testing or introspection only.
+        # actor_class = staticmethod(make_actor_class)
+
+    return StageProxy

nv_ingest/version.py

@@ -5,7 +5,6 @@
 
 import datetime
 import os
-import re
 
 
 def get_version():
@@ -16,13 +15,6 @@ def get_version():
     if not version:
         version = f"{datetime.datetime.now().strftime('%Y.%m.%d')}"
 
-    # We only check this for dev, we assume for release the user knows what they are doing
-    if release_type != "release":
-        # Ensure the version is PEP 440 compatible
-        pep440_regex = r"^\d{4}\.\d{1,2}\.\d{1,2}$"
-        if not re.match(pep440_regex, version):
-            raise ValueError(f"Version '{version}' is not PEP 440 compatible")
-
     # Construct the final version string
     if release_type == "dev":
         # If rev is not specified and defaults to 0 lets create a more meaningful development

{nv_ingest-2025.6.2.dev20250602.dist-info → nv_ingest-2025.7.7.dev20250707.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 2025.6.2.dev20250602
+Version: 2025.7.7.dev20250707
 Summary: Python module for multimodal document ingestion
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License
@@ -240,13 +240,13 @@ Requires-Dist: python-docx>=1.1.2
 Requires-Dist: python-dotenv>=1.0.1
 Requires-Dist: python-pptx>=1.0.2
 Requires-Dist: prometheus-client
-Requires-Dist: torch==2.4.1
+Requires-Dist: torch>=2.4.1
 Requires-Dist: ray[all]>=2.37.0
 Requires-Dist: redis>=5.2.1
 Requires-Dist: requests>=2.28.2
 Requires-Dist: scikit-learn>=1.6.0
 Requires-Dist: scipy>=1.15.1
-Requires-Dist: setuptools>=58.2.0
+Requires-Dist: setuptools>=78.1.1
 Requires-Dist: tabulate>=0.9.0
 Requires-Dist: torchvision
 Requires-Dist: torchaudio
@@ -259,7 +259,7 @@ Requires-Dist: opencv-python
 Requires-Dist: pymilvus>=2.5.10
 Requires-Dist: pymilvus[bulk_writer,model]
 Requires-Dist: tritonclient
-Requires-Dist: nvidia-riva-client>=2.18.0
+Requires-Dist: nvidia-riva-client==2.20.0
 Requires-Dist: unstructured-client
 Requires-Dist: markitdown
 Dynamic: license-file

{nv_ingest-2025.6.2.dev20250602.dist-info → nv_ingest-2025.7.7.dev20250707.dist-info}/RECORD

@@ -1,5 +1,5 @@
 nv_ingest/__init__.py,sha256=vJLPeuxiIHqbxXPJSu9qe3MS-GPavbOUExyRq83DxxM,895
-nv_ingest/version.py,sha256=Y9gMjlV_tnRSE3JbmS1rWIfVppM974_g0k30MRF3IQM,1352
+nv_ingest/version.py,sha256=MG7DxlzpnoJI56vqxwzs9WeMAEI3uPhfDiNLs6GN6wI,986
 nv_ingest/api/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/api/main.py,sha256=XE-p4lJp1E7CCDOB8ENtYFrf63Dtq2bzQiGxpRfL2LA,1603
 nv_ingest/api/v1/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
@@ -14,15 +14,15 @@ nv_ingest/framework/orchestration/ray/edges/async_queue_edge.py,sha256=PQliU_kyG
 nv_ingest/framework/orchestration/ray/edges/ray_queue_edge.py,sha256=VFii2yxJuikimOxie3edKq5JN06g78AF8bdHSHVX8p8,2677
 nv_ingest/framework/orchestration/ray/edges/threaded_queue_edge.py,sha256=N6NH4KgZJ60e_JkGRcSmfQtX37qtX4TMcavOR-n3heE,2549
 nv_ingest/framework/orchestration/ray/examples/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/framework/orchestration/ray/examples/pipeline_test_harness.py,sha256=QHhoHjELAjs1SNGUOcUBU7f9JXxq3xTwM6fHyzcQytg,16450
+nv_ingest/framework/orchestration/ray/examples/pipeline_test_harness.py,sha256=qUNvWiNBUEEzuCySY3usWmHQz9qMgTGVZuKmLWqTsi4,16412
 nv_ingest/framework/orchestration/ray/examples/task_source_harness.py,sha256=Yt7uxThg7s8WuMiaHLKC8r1XAG7QixegfkT-juE5oNw,1953
 nv_ingest/framework/orchestration/ray/examples/task_source_sink_harness.py,sha256=XkvsoIzH5ftXvAZ4ox7mxbx7ESVx6D8Xupcwbqgd52w,3277
 nv_ingest/framework/orchestration/ray/primitives/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/primitives/dataclasses.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 nv_ingest/framework/orchestration/ray/primitives/pipeline_monitor.py,sha256=L8ENPiF-lxqhIXVEQwQD5CCqQMb710ynj5D_Y4ixGhs,11077
-nv_ingest/framework/orchestration/ray/primitives/pipeline_topology.py,sha256=gc9gZNqPmnP76M-u8sQXyJd5aTSlyY_0CjLYNa-zvzk,29106
-nv_ingest/framework/orchestration/ray/primitives/ray_pipeline.py,sha256=BEBLjkYFXIH396EUQcfuxhrWlIMs9i6z7YfeeqJ5cZg,59579
-nv_ingest/framework/orchestration/ray/primitives/ray_stat_collector.py,sha256=yPIvOhxY42P-gf5dLkcPkfvfwL_I-ay0C8k5eNaU-VA,15811
+nv_ingest/framework/orchestration/ray/primitives/pipeline_topology.py,sha256=2Xg7QoKKPPFUWkLck7NtEtb1xLnK3b5uUw8LRxPhLyw,29106
+nv_ingest/framework/orchestration/ray/primitives/ray_pipeline.py,sha256=6i0EGWZ9WXpPVkfLwP5a2Y45gwAhQjWjobTp_kuFPsE,60478
+nv_ingest/framework/orchestration/ray/primitives/ray_stat_collector.py,sha256=AJ79OTh_NxxoTcyBNiopq3K_nLumsB9UU_axqQS3Gus,15810
 nv_ingest/framework/orchestration/ray/stages/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/extractors/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/extractors/audio_extractor.py,sha256=KV4hvY0NTGG8CjZviTgcFLQzaH8WJJGkkb9PFYbROww,3417
@@ -35,11 +35,11 @@ nv_ingest/framework/orchestration/ray/stages/extractors/pdf_extractor.py,sha256=
 nv_ingest/framework/orchestration/ray/stages/extractors/pptx_extractor.py,sha256=ywPGA-3GNsbp3FWFsu04foumM6ZCccRrm73ijS7oY0g,3581
 nv_ingest/framework/orchestration/ray/stages/extractors/table_extractor.py,sha256=EOcjyJYAB3TuXewZFld4shnGQUQ9VysjPrIWnmb8zuI,3893
 nv_ingest/framework/orchestration/ray/stages/injectors/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/framework/orchestration/ray/stages/injectors/metadata_injector.py,sha256=cdGbLBH0x-4uCdhr6JH_dMVFyBqPODQ5_WYO1otk8tI,4147
+nv_ingest/framework/orchestration/ray/stages/injectors/metadata_injector.py,sha256=K8jase7PD9kd8AuntzjdS1IO4ae8Oo_6byZsFG777D0,6838
 nv_ingest/framework/orchestration/ray/stages/meta/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_edge_base.py,sha256=LnVqBJmpfCmcI-eJLbkwK-7SS-hpEp98P4iCRv_Zhb0,1726
 nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_sink_stage_base.py,sha256=AhlZUbDK2Jckqnu8hVbJrckW8MsSixfmWc1bst9gRYk,3447
-nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_source_stage_base.py,sha256=MtePbSiouDOtNYQ8bQI3hkHSXqkjN9r-NU2lXTz8paM,1793
+nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_source_stage_base.py,sha256=1Pae2xRPK0_QLh53yHECVFm2guwgvZaiRRr3tp4OpYI,1744
 nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_stage_base.py,sha256=rAuEH8uq8-j4Ipkb1zMB8z_x_PMvxwO9LFN4iY7UXjE,28957
 nv_ingest/framework/orchestration/ray/stages/mutate/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/mutate/image_dedup.py,sha256=UepeDvH6Cfgm5rIylRx6uOxihS0OZ4Q1DGUrjUybNaY,3493
@@ -48,7 +48,7 @@ nv_ingest/framework/orchestration/ray/stages/sinks/__init__.py,sha256=wQSlVx3T14
 nv_ingest/framework/orchestration/ray/stages/sinks/default_drain.py,sha256=0SQHJlFuXlP16YRWduX1fMKgjhUd7UhDAWQ8XZh4_0I,1471
 nv_ingest/framework/orchestration/ray/stages/sinks/message_broker_task_sink.py,sha256=enylryvcPmzirpOjCahqYJbNSLsNvv1KpMnOzGqNZQQ,11509
 nv_ingest/framework/orchestration/ray/stages/sources/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
-nv_ingest/framework/orchestration/ray/stages/sources/message_broker_task_source.py,sha256=srDsgp8ExMHZNI76ch3iX7S0drMXmQ3NkWC_udnwqmo,20286
+nv_ingest/framework/orchestration/ray/stages/sources/message_broker_task_source.py,sha256=9YoVytbFFt-RpIR_MN2m3T93zVTjts8tjhi0qzLJkTw,19922
 nv_ingest/framework/orchestration/ray/stages/storage/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/storage/image_storage.py,sha256=6NkwQzseAnaj0Ptpr3oKvab2EnJdMwTjI2p4dS_HzsI,3901
 nv_ingest/framework/orchestration/ray/stages/storage/store_embeddings.py,sha256=SMLHQElZkKldnjy0_VHIKS65DBAAtOhwhdoaFe1yb9I,3337
@@ -58,16 +58,17 @@ nv_ingest/framework/orchestration/ray/stages/telemetry/otel_meter.py,sha256=wQSl
 nv_ingest/framework/orchestration/ray/stages/telemetry/otel_tracer.py,sha256=53M8Xw84gcuRAeHbf4Z_ApLFX9Zkd1KSec_2k-wXl6c,7947
 nv_ingest/framework/orchestration/ray/stages/transforms/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/transforms/image_caption.py,sha256=OuqPmJmCqbg9k7roDivuvfYVTd05Nl9PMC0_E9PHgYw,3514
-nv_ingest/framework/orchestration/ray/stages/transforms/text_embed.py,sha256=Hh7UXVZSS5LKt3uU5gwWIQ7SYqQrSN9tzFYw4CeUpUA,3535
-nv_ingest/framework/orchestration/ray/stages/transforms/text_splitter.py,sha256=ulFDsTuz4oivvm_FKACfN1KH9X33mGH87LfD7rkaJnY,3090
+nv_ingest/framework/orchestration/ray/stages/transforms/text_embed.py,sha256=cERKHTMkfOLGpibApCpGC_8URoIi-wn-Ttv-4b2XpJ0,3556
+nv_ingest/framework/orchestration/ray/stages/transforms/text_splitter.py,sha256=hyTIv3MMnuyZmgbBHbXicSgk9DhbpygDrgUtXRwvlmo,4677
 nv_ingest/framework/orchestration/ray/stages/utility/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/stages/utility/throughput_monitor.py,sha256=MB27CkoNeuirN6CUHgjsC5Wh958NF7m_N7HE4VKfx3k,2264
 nv_ingest/framework/orchestration/ray/util/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/util/pipeline/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/util/pipeline/pid_controller.py,sha256=AWyCFPP41vp1NOkO2urqm7vh-sTGKypJxwhdq8HxK6Q,50681
-nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_builders.py,sha256=jMYnVe_0rb1OIO9mlB4LH3uXtgaXBbUG-rDPx6fe6J8,10456
-nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_runners.py,sha256=3aSYSxyunm-eKUYErDArQTHXSoNKlNJMUr9o5Ui6VTk,14037
-nv_ingest/framework/orchestration/ray/util/pipeline/stage_builders.py,sha256=_MPUbOVTo9CjkBdDA--mcpu2plQ9qFY_TCBXbfpbB_A,21477
+nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_builders.py,sha256=d2-GS2tqk6JOFdw65CL1AwfjdUbkC_XxUuJH8Dy-aQ0,10456
+nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_runners.py,sha256=C7wf0AdsOpfYUTMKWLhi8hkoUnnCAD8v3OMwOkpYgKw,14331
+nv_ingest/framework/orchestration/ray/util/pipeline/stage_builders.py,sha256=sEpfVgZrnnM_kZ5KoSBMufO6iU4Z8v8XskMbOGumg2g,21415
+nv_ingest/framework/orchestration/ray/util/pipeline/tools.py,sha256=LQVb8k9jURaxh2Ga44Js_XuYFCbeN4_nLgDmtExovQg,8026
 nv_ingest/framework/orchestration/ray/util/system_tools/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/orchestration/ray/util/system_tools/memory.py,sha256=ICqY0LLB3hFTZk03iX5yffMSKFH2q_aQomtDVzS_mKw,2228
 nv_ingest/framework/orchestration/ray/util/system_tools/visualizers.py,sha256=2oHZdO_3L1LGuzpyNmZBDh19n0E-APAaHk4MEwBwSHs,12895
@@ -95,8 +96,8 @@ nv_ingest/framework/util/service/meta/ingest/__init__.py,sha256=wQSlVx3T14ZgQAt-
 nv_ingest/framework/util/service/meta/ingest/ingest_service_meta.py,sha256=QS3uNxWBl5dIcmIpJKNe8_TLcTUuN2vcKyHeAwa-eSo,1589
 nv_ingest/framework/util/telemetry/__init__.py,sha256=wQSlVx3T14ZgQAt-EPzEczQusXVW0W8yynnUaFFGE3s,143
 nv_ingest/framework/util/telemetry/global_stats.py,sha256=nq65pEEdiwjAfGiqsxG1CeQMC96O3CfQxsZuGFCY-ds,4554
-nv_ingest-2025.6.2.dev20250602.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-nv_ingest-2025.6.2.dev20250602.dist-info/METADATA,sha256=PiLNsp7gjUH7ll310hQOEQCNk6i1Bf-7D7kg38MnLps,15141
-nv_ingest-2025.6.2.dev20250602.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-nv_ingest-2025.6.2.dev20250602.dist-info/top_level.txt,sha256=sjb0ajIsgn3YgftSjZHlYO0HjYAIIhNuXG_AmywCvaU,10
-nv_ingest-2025.6.2.dev20250602.dist-info/RECORD,,
+nv_ingest-2025.7.7.dev20250707.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+nv_ingest-2025.7.7.dev20250707.dist-info/METADATA,sha256=TmUXlozq6Ohz6qHFz7CTBrqa1Pwwxum9bMm6EblzLxU,15141
+nv_ingest-2025.7.7.dev20250707.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+nv_ingest-2025.7.7.dev20250707.dist-info/top_level.txt,sha256=sjb0ajIsgn3YgftSjZHlYO0HjYAIIhNuXG_AmywCvaU,10
+nv_ingest-2025.7.7.dev20250707.dist-info/RECORD,,