nv-ingest 2025.12.12.dev20251212__tar.gz → 2025.12.20.dev20251220__tar.gz

{nv_ingest-2025.12.12.dev20251212 → nv_ingest-2025.12.20.dev20251220}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 2025.12.12.dev20251212
+Version: 2025.12.20.dev20251220
 Summary: Python module for multimodal document ingestion
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License

{nv_ingest-2025.12.12.dev20251212 → nv_ingest-2025.12.20.dev20251220}/nv_ingest/api/main.py

@@ -23,7 +23,7 @@ logger = logging.getLogger(__name__)
 app = FastAPI(
     title="NV-Ingest Microservice",
     description="Service for ingesting heterogenous datatypes",
-    version="25.4.2",
+    version="26.1.0",
     contact={
         "name": "NVIDIA Corporation",
         "url": "https://nvidia.com",

nv_ingest-2025.12.20.dev20251220/nv_ingest/framework/orchestration/ray/stages/meta/udf_parallel_helper.py

@@ -0,0 +1,64 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024-25, NVIDIA CORPORATION & AFFILIATES.
+# All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+# Added this no-op UDF ray stage to the pipeline to help speed up the LLM api calls
+
+"""
+UDF Parallel Stage - A high-concurrency no-op stage for parallel UDF execution.
+
+This stage does nothing except pass messages through, but with high replica count
+it provides a parallel execution pool for UDFs to achieve N-way concurrency.
+"""
+
+import logging
+from typing import Any, Optional
+from pydantic import BaseModel
+import ray
+
+from nv_ingest.framework.orchestration.ray.stages.meta.ray_actor_stage_base import RayActorStage
+from nv_ingest.framework.util.flow_control.udf_intercept import udf_intercept_hook
+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__)
+
+
+@ray.remote
+class UDFParallelStage(RayActorStage):
+    """
+    A no-op pass-through stage designed for parallel UDF execution.
+
+    This stage simply returns the input message unchanged, but when configured
+    with multiple replicas, it provides a high-concurrency pool for UDFs to
+    achieve parallel execution without blocking.
+    """
+
+    def __init__(self, config: BaseModel, stage_name: Optional[str] = None) -> None:
+        super().__init__(config, stage_name=stage_name)
+        logger.info(f"UDFParallelStage initialized: {stage_name}")
+
+    @nv_ingest_node_failure_try_except()
+    @traceable()
+    @udf_intercept_hook()
+    def on_data(self, message: Any) -> Any:
+        """
+        Pass-through processing that simply returns the message unchanged.
+
+        The @udf_intercept_hook decorator allows UDFs to target this stage,
+        and multiple replicas provide parallel execution capacity.
+
+        Parameters
+        ----------
+        message : Any
+            The incoming control message.
+
+        Returns
+        -------
+        Any
+            The unmodified control message.
+        """
+        # No-op: just return the message
+        return message

{nv_ingest-2025.12.12.dev20251212 → nv_ingest-2025.12.20.dev20251220}/nv_ingest.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 2025.12.12.dev20251212
+Version: 2025.12.20.dev20251220
 Summary: Python module for multimodal document ingestion
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License

{nv_ingest-2025.12.12.dev20251212 → nv_ingest-2025.12.20.dev20251220}/nv_ingest.egg-info/SOURCES.txt

@@ -63,6 +63,7 @@ nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_edge_base.py
 nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_sink_stage_base.py
 nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_source_stage_base.py
 nv_ingest/framework/orchestration/ray/stages/meta/ray_actor_stage_base.py
+nv_ingest/framework/orchestration/ray/stages/meta/udf_parallel_helper.py
 nv_ingest/framework/orchestration/ray/stages/mutate/__init__.py
 nv_ingest/framework/orchestration/ray/stages/mutate/image_dedup.py
 nv_ingest/framework/orchestration/ray/stages/mutate/image_filter.py