nv-ingest 2025.12.8.dev20251208__tar.gz → 2025.12.12.dev20251212__tar.gz

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 2025.12.8.dev20251208
+Version: 2025.12.12.dev20251212
 Summary: Python module for multimodal document ingestion
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/nv_ingest/api/v2/ingest.py

@@ -1003,7 +1003,7 @@ async def submit_job_v2(
                         "page_count": chunk.get("page_count", 0),
                     }
                 )
-            logger.error(f"Removing uploaded file {upload_path}")
+            logger.debug(f"Removing uploaded file {upload_path}")
             os.remove(upload_path)
 
         if submission_items:

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/nv_ingest/framework/orchestration/process/execution.py

@@ -162,6 +162,11 @@ def build_logging_config_from_env() -> LoggingConfig:
         if key not in os.environ:
             os.environ[key] = default_value
 
+    # For PRODUCTION mode, also suppress nv-ingest module INFO logs
+    if preset_level == "PRODUCTION":
+        logging.getLogger("nv_ingest").setLevel(logging.WARNING)
+        logging.getLogger("nv_ingest_api").setLevel(logging.WARNING)
+
     logger.info(f"Applied Ray logging preset: {preset_level}")
 
     # Get log level from environment, default to INFO
@@ -324,6 +329,7 @@ def launch_pipeline(
     pipeline_config = resolve_static_replicas(pipeline_config)
 
     # Pretty print the final pipeline configuration (after replica resolution)
+    # INFO level so it shows in docker/helm deployments; quiet mode suppresses in library mode
     pretty_output = pretty_print_pipeline_config(pipeline_config, config_path=None)
     logger.info("\n" + pretty_output)
 

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/nv_ingest/framework/orchestration/ray/examples/pipeline_test_harness.py

@@ -150,7 +150,7 @@ if __name__ == "__main__":
     os.environ["OCR_GRPC_ENDPOINT"] = "localhost:8010"
     os.environ["OCR_INFER_PROTOCOL"] = "grpc"
     os.environ["OCR_MODEL_NAME"] = "paddle"
-    os.environ["NEMORETRIEVER_PARSE_HTTP_ENDPOINT"] = "https://integrate.api.nvidia.com/v1/chat/completions"
+    os.environ["NEMOTRON_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"] = "nvidia/nemotron-nano-12b-v2-vl"
     logger.info("Environment variables set.")
@@ -170,23 +170,23 @@ if __name__ == "__main__":
         yolox_graphic_elements_auth,
         yolox_graphic_elements_protocol,
     ) = get_nim_service("yolox_graphic_elements")
-    nemoretriever_parse_grpc, nemoretriever_parse_http, nemoretriever_parse_auth, nemoretriever_parse_protocol = (
-        get_nim_service("nemoretriever_parse")
+    nemotron_parse_grpc, nemotron_parse_http, nemotron_parse_auth, nemotron_parse_protocol = get_nim_service(
+        "nemotron_parse"
     )
     ocr_grpc, ocr_http, ocr_auth, ocr_protocol = get_nim_service("ocr")
 
-    model_name = os.environ.get("NEMORETRIEVER_PARSE_MODEL_NAME", "nvidia/nemoretriever-parse")
+    model_name = os.environ.get("NEMOTRON_PARSE_MODEL_NAME", "nvidia/nemotron-parse")
     pdf_extractor_config = {
         "pdfium_config": {
             "auth_token": yolox_auth,  # All auth tokens are the same for the moment
             "yolox_endpoints": (yolox_grpc, yolox_http),
             "yolox_infer_protocol": yolox_protocol,
         },
-        "nemoretriever_parse_config": {
-            "auth_token": nemoretriever_parse_auth,
-            "nemoretriever_parse_endpoints": (nemoretriever_parse_grpc, nemoretriever_parse_http),
-            "nemoretriever_parse_infer_protocol": nemoretriever_parse_protocol,
-            "nemoretriever_parse_model_name": model_name,
+        "nemotron_parse_config": {
+            "auth_token": nemotron_parse_auth,
+            "nemotron_parse_endpoints": (nemotron_parse_grpc, nemotron_parse_http),
+            "nemotron_parse_infer_protocol": nemotron_parse_protocol,
+            "nemotron_parse_model_name": model_name,
             "yolox_endpoints": (yolox_grpc, yolox_http),
             "yolox_infer_protocol": yolox_protocol,
         },

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_runners.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import logging
+import os
 from typing import Union, Optional, TextIO
 
 
@@ -23,6 +24,34 @@ from nv_ingest.framework.orchestration.execution.helpers import (
 logger = logging.getLogger(__name__)
 
 
+def _configure_quiet_mode():
+    """
+    Configure environment for quiet/production logging in library mode.
+
+    Sets INGEST_RAY_LOG_LEVEL=PRODUCTION if not already set by user, which:
+    - Sets Ray logging to ERROR level (suppresses INFO/WARNING)
+    - Disables Ray usage stats collection
+    - Disables Ray import warnings
+
+    Also silences other common warnings that are noisy in library mode.
+    """
+    # Only set if user hasn't explicitly configured
+    if "INGEST_RAY_LOG_LEVEL" not in os.environ:
+        os.environ["INGEST_RAY_LOG_LEVEL"] = "PRODUCTION"
+
+    # Silence Ray accelerator env var warning
+    if "RAY_ACCEL_ENV_VAR_OVERRIDE_ON_ZERO" not in os.environ:
+        os.environ["RAY_ACCEL_ENV_VAR_OVERRIDE_ON_ZERO"] = "0"
+
+    # Disable OTEL tracing export errors (no collector expected in library mode)
+    if "OTEL_SDK_DISABLED" not in os.environ:
+        os.environ["OTEL_SDK_DISABLED"] = "true"
+
+    # Set nv-ingest module loggers to WARNING to suppress INFO level startup messages
+    logging.getLogger("nv_ingest").setLevel(logging.WARNING)
+    logging.getLogger("nv_ingest_api").setLevel(logging.WARNING)
+
+
 def run_pipeline(
     pipeline_config: Optional[PipelineConfigSchema] = None,
     block: bool = True,
@@ -32,6 +61,7 @@ def run_pipeline(
     stdout: Optional[TextIO] = None,
     stderr: Optional[TextIO] = None,
     libmode: bool = True,
+    quiet: Optional[bool] = None,
 ) -> Union[RayPipelineInterface, float, RayPipelineSubprocessInterface]:
     """
     Launch and manage a pipeline using configuration.
@@ -65,6 +95,10 @@ def run_pipeline(
     libmode : bool, default=True
         If True and pipeline_config is None, loads the default libmode pipeline configuration.
         If False, requires pipeline_config to be provided.
+    quiet : Optional[bool], default=None
+        If True, configures logging for minimal output (PRODUCTION preset, suppresses
+        INFO-level startup messages). If None, defaults to True when libmode=True.
+        Set to False to see verbose startup logs even in library mode.
 
     Returns
     -------
@@ -83,6 +117,12 @@ def run_pipeline(
     Exception
         Any other exceptions raised during pipeline launch or configuration.
     """
+    # Configure quiet mode for library mode by default (unless explicitly disabled)
+    if quiet is None:
+        quiet = libmode
+    if quiet:
+        _configure_quiet_mode()
+
     # Resolve configuration
     config = resolve_pipeline_config(pipeline_config, libmode)
     overrides = create_runtime_overrides(disable_dynamic_scaling, dynamic_memory_threshold)

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/nv_ingest/pipeline/config/replica_resolver.py

@@ -11,6 +11,7 @@ consumption stays within the static_memory_threshold.
 """
 
 import logging
+import os
 from typing import List
 from copy import deepcopy
 
@@ -102,8 +103,17 @@ def resolve_static_replicas(pipeline_config: PipelineConfigSchema) -> PipelineCo
 
     logger.info(f"Total baseline memory demand: {total_memory_demand_mb}MB from {len(non_static_stages)} stages")
 
-    # Check if we need to scale down
-    if total_memory_demand_mb <= available_memory_mb:
+    # Optional bypass of global memory-based scale down via environment variable
+    bypass_env = os.getenv("NV_INGEST_BYPASS_STATIC_MEMORY_SCALE_DOWN", "").strip().lower()
+    bypass_scale_down = bypass_env in ("1", "true", "yes", "on")
+
+    # Check if we need to scale down (unless bypassed)
+    if bypass_scale_down:
+        logger.warning(
+            "Bypassing static memory-based replica scale-down due to NV_INGEST_BYPASS_STATIC_MEMORY_SCALE_DOWN"
+        )
+        scaling_factor = 1.0
+    elif total_memory_demand_mb <= available_memory_mb:
         logger.info("Memory demand within threshold, applying baseline replica counts")
         scaling_factor = 1.0
     else:

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/nv_ingest/pipeline/default_libmode_pipeline_impl.py

@@ -71,14 +71,14 @@ stages:
           $YOLOX_HTTP_ENDPOINT|"https://ai.api.nvidia.com/v1/cv/nvidia/nemoretriever-page-elements-v2"
         ]
         yolox_infer_protocol: $YOLOX_INFER_PROTOCOL|http
-      nemoretriever_parse_config:
+      nemotron_parse_config:
         auth_token: $NGC_API_KEY|$NVIDIA_API_KEY
-        nemoretriever_parse_endpoints: [
-          $NEMORETRIEVER_PARSE_GRPC_ENDPOINT|"",
-          $NEMORETRIEVER_PARSE_HTTP_ENDPOINT|"https://integrate.api.nvidia.com/v1/chat/completions"
+        nemotron_parse_endpoints: [
+          $NEMOTRON_PARSE_GRPC_ENDPOINT|"",
+          $NEMOTRON_PARSE_HTTP_ENDPOINT|"https://integrate.api.nvidia.com/v1/chat/completions"
         ]
-        nemoretriever_parse_infer_protocol: $NEMORETRIEVER_PARSE_INFER_PROTOCOL|http
-        nemoretriever_parse_model_name: $NEMORETRIEVER_PARSE_MODEL_NAME|"nvidia/nemoretriever-parse"
+        nemotron_parse_infer_protocol: $NEMOTRON_PARSE_INFER_PROTOCOL|http
+        nemotron_parse_model_name: $NEMOTRON_PARSE_MODEL_NAME|"nvidia/nemotron-parse"
         yolox_endpoints: [
           $YOLOX_GRPC_ENDPOINT|"",
           $YOLOX_HTTP_ENDPOINT|"https://ai.api.nvidia.com/v1/cv/nvidia/nemoretriever-page-elements-v2"
@@ -334,7 +334,8 @@ stages:
       api_key: $NGC_API_KEY|$NVIDIA_API_KEY
       endpoint_url: $VLM_CAPTION_ENDPOINT|"http://vlm:8000/v1/chat/completions"
       model_name: $VLM_CAPTION_MODEL_NAME|"nvidia/nemotron-nano-12b-v2-vl"
-      prompt: "Caption the content of this image:"
+      prompt: $VLM_CAPTION_PROMPT|"Caption the content of this image:"
+      system_prompt: $VLM_CAPTION_SYSTEM_PROMPT|"/no_think"
     replicas:
       min_replicas: 0
       max_replicas:

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/nv_ingest/pipeline/default_pipeline_impl.py

@@ -70,14 +70,14 @@ stages:
           $YOLOX_HTTP_ENDPOINT|"http://page-elements:8000/v1/infer",
         ]
         yolox_infer_protocol: $YOLOX_INFER_PROTOCOL|grpc
-      nemoretriever_parse_config:
+      nemotron_parse_config:
         auth_token: $NGC_API_KEY|$NVIDIA_API_KEY
-        nemoretriever_parse_endpoints: [
-          $NEMORETRIEVER_PARSE_GRPC_ENDPOINT|"",
-          $NEMORETRIEVER_PARSE_HTTP_ENDPOINT|"http://nemoretriever-parse:8000/v1/chat/completions",
+        nemotron_parse_endpoints: [
+          $NEMOTRON_PARSE_GRPC_ENDPOINT|"",
+          $NEMOTRON_PARSE_HTTP_ENDPOINT|"http://nemotron-parse:8000/v1/chat/completions",
         ]
-        nemoretriever_parse_infer_protocol: $NEMORETRIEVER_PARSE_INFER_PROTOCOL|http
-        nemoretriever_parse_model_name: $NEMORETRIEVER_PARSE_MODEL_NAME|"nvidia/nemoretriever-parse"
+        nemotron_parse_infer_protocol: $NEMOTRON_PARSE_INFER_PROTOCOL|http
+        nemotron_parse_model_name: $NEMOTRON_PARSE_MODEL_NAME|"nvidia/nemotron-parse"
         yolox_endpoints: [
           $YOLOX_GRPC_ENDPOINT|"page-elements:8001",
           $YOLOX_HTTP_ENDPOINT|"http://page-elements:8000/v1/infer",
@@ -354,7 +354,8 @@ stages:
       api_key: $NGC_API_KEY|$NVIDIA_API_KEY
       model_name: $VLM_CAPTION_MODEL_NAME|"nvidia/nemotron-nano-12b-v2-vl"
       endpoint_url: $VLM_CAPTION_ENDPOINT|"http://vlm:8000/v1/chat/completions"
-      prompt: "Caption the content of this image:"
+      prompt: $VLM_CAPTION_PROMPT|"Caption the content of this image:"
+      system_prompt: $VLM_CAPTION_SYSTEM_PROMPT|"/no_think"
     replicas:
       min_replicas: 0
       max_replicas:

{nv_ingest-2025.12.8.dev20251208 → nv_ingest-2025.12.12.dev20251212}/nv_ingest.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 2025.12.8.dev20251208
+Version: 2025.12.12.dev20251212
 Summary: Python module for multimodal document ingestion
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License