nv-ingest 2025.7.22.dev20250722__tar.gz → 2025.7.24.dev20250724__tar.gz

{nv_ingest-2025.7.22.dev20250722 → nv_ingest-2025.7.24.dev20250724}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 2025.7.22.dev20250722
+Version: 2025.7.24.dev20250724
 Summary: Python module for multimodal document ingestion
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License
@@ -231,7 +231,7 @@ Requires-Dist: opentelemetry-exporter-otlp>=1.27.0
 Requires-Dist: opentelemetry-sdk>=1.27.0
 Requires-Dist: pydantic>2.0.0
 Requires-Dist: pydantic-settings>2.0.0
-Requires-Dist: pypdfium2==4.30.1
+Requires-Dist: pypdfium2==4.30.0
 Requires-Dist: pytest>=8.0.2
 Requires-Dist: pytest-mock>=3.14.0
 Requires-Dist: pytest-cov>=6.0.0
@@ -240,7 +240,6 @@ 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: ray[all]>=2.37.0
 Requires-Dist: redis>=5.2.1
 Requires-Dist: requests>=2.28.2
@@ -248,8 +247,6 @@ Requires-Dist: scikit-learn>=1.6.0
 Requires-Dist: scipy>=1.15.1
 Requires-Dist: setuptools>=78.1.1
 Requires-Dist: tabulate>=0.9.0
-Requires-Dist: torchvision
-Requires-Dist: torchaudio
 Requires-Dist: transformers>=4.47.0
 Requires-Dist: tqdm>=4.67.1
 Requires-Dist: uvicorn

nv_ingest-2025.7.24.dev20250724/nv_ingest/framework/orchestration/ray/util/env_config.py

@@ -0,0 +1,75 @@
+import os
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+def str_to_bool(value: str) -> bool:
+    """
+    Convert string to boolean value.
+
+    Parameters
+    ----------
+    value : str
+        String value to convert
+
+    Returns
+    -------
+    bool
+        Boolean representation of the string
+    """
+    return value.strip().lower() in {"1", "true", "yes", "on"}
+
+
+def get_env_var(name: str, default, var_type=None):
+    """
+    Get environment variable with type conversion and default value.
+
+    Parameters
+    ----------
+    name : str
+        Environment variable name
+    default : Any
+        Default value if environment variable is not set
+    var_type : type, optional
+        Type to convert to. If None, infers from default value type
+
+    Returns
+    -------
+    Any
+        Environment variable value converted to the appropriate type
+    """
+    value = os.environ.get(name)
+    if value is None:
+        return default
+
+    # Determine type from default if not explicitly provided
+    target_type = var_type or type(default)
+
+    # Handle boolean conversion specially
+    if target_type is bool:
+        return str_to_bool(value)
+
+    # For other types, use direct conversion
+    try:
+        return target_type(value)
+    except (ValueError, TypeError) as e:
+        logger.warning(
+            f"Failed to convert environment variable {name}='{value}' to \
+                  {target_type.__name__}. Using default: {default}, error: {e}"
+        )
+        return default
+
+
+# Dynamic Memory Scaling Configuration
+DISABLE_DYNAMIC_SCALING = get_env_var("INGEST_DISABLE_DYNAMIC_SCALING", False, bool)
+DYNAMIC_MEMORY_THRESHOLD = get_env_var("INGEST_DYNAMIC_MEMORY_THRESHOLD", 0.75, float)
+DYNAMIC_MEMORY_KP = get_env_var("INGEST_DYNAMIC_MEMORY_KP", 0.2, float)
+DYNAMIC_MEMORY_KI = get_env_var("INGEST_DYNAMIC_MEMORY_KI", 0.01, float)
+DYNAMIC_MEMORY_EMA_ALPHA = get_env_var("INGEST_DYNAMIC_MEMORY_EMA_ALPHA", 0.1, float)
+DYNAMIC_MEMORY_TARGET_QUEUE_DEPTH = get_env_var("INGEST_DYNAMIC_MEMORY_TARGET_QUEUE_DEPTH", 0, int)
+DYNAMIC_MEMORY_PENALTY_FACTOR = get_env_var("INGEST_DYNAMIC_MEMORY_PENALTY_FACTOR", 0.1, float)
+DYNAMIC_MEMORY_ERROR_BOOST_FACTOR = get_env_var("INGEST_DYNAMIC_MEMORY_ERROR_BOOST_FACTOR", 1.5, float)
+DYNAMIC_MEMORY_RCM_MEMORY_SAFETY_BUFFER_FRACTION = get_env_var(
+    "INGEST_DYNAMIC_MEMORY_RCM_MEMORY_SAFETY_BUFFER_FRACTION", 0.15, float
+)

{nv_ingest-2025.7.22.dev20250722 → nv_ingest-2025.7.24.dev20250724}/nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_runners.py

@@ -23,18 +23,21 @@ from nv_ingest.framework.orchestration.ray.primitives.ray_pipeline import (
     RayPipelineInterface,
 )
 from nv_ingest.framework.orchestration.ray.util.pipeline.pipeline_builders import setup_ingestion_pipeline
+from nv_ingest.framework.orchestration.ray.util.env_config import (
+    DISABLE_DYNAMIC_SCALING,
+    DYNAMIC_MEMORY_THRESHOLD,
+    DYNAMIC_MEMORY_KP,
+    DYNAMIC_MEMORY_KI,
+    DYNAMIC_MEMORY_EMA_ALPHA,
+    DYNAMIC_MEMORY_TARGET_QUEUE_DEPTH,
+    DYNAMIC_MEMORY_PENALTY_FACTOR,
+    DYNAMIC_MEMORY_ERROR_BOOST_FACTOR,
+    DYNAMIC_MEMORY_RCM_MEMORY_SAFETY_BUFFER_FRACTION,
+)
 
 logger = logging.getLogger(__name__)
 
 
-def str_to_bool(value: str) -> bool:
-    return value.strip().lower() in {"1", "true", "yes", "on"}
-
-
-DISABLE_DYNAMIC_SCALING = str_to_bool(os.environ.get("INGEST_DISABLE_DYNAMIC_SCALING", "false"))
-DYNAMIC_MEMORY_THRESHOLD = float(os.environ.get("INGEST_DYNAMIC_MEMORY_THRESHOLD", 0.75))
-
-
 class PipelineCreationSchema(BaseModel):
     """
     Schema for pipeline creation configuration.
@@ -88,7 +91,7 @@ class PipelineCreationSchema(BaseModel):
     # Vision language model settings
     vlm_caption_endpoint: str = os.getenv(
         "VLM_CAPTION_ENDPOINT",
-        "https://ai.api.nvidia.com/v1/gr/nvidia/llama-3.1-nemotron-nano-vl-8b-v1/chat/completions",
+        "https://integrate.api.nvidia.com/v1/chat/completions",
     )
     vlm_caption_model_name: str = os.getenv("VLM_CAPTION_MODEL_NAME", "nvidia/llama-3.1-nemotron-nano-vl-8b-v1")
 
@@ -235,7 +238,15 @@ def _launch_pipeline(
     dynamic_memory_threshold = dynamic_memory_threshold if dynamic_memory_threshold else DYNAMIC_MEMORY_THRESHOLD
 
     scaling_config = ScalingConfig(
-        dynamic_memory_scaling=dynamic_memory_scaling, dynamic_memory_threshold=dynamic_memory_threshold
+        dynamic_memory_scaling=dynamic_memory_scaling,
+        dynamic_memory_threshold=dynamic_memory_threshold,
+        pid_kp=DYNAMIC_MEMORY_KP,
+        pid_ki=DYNAMIC_MEMORY_KI,
+        pid_ema_alpha=DYNAMIC_MEMORY_EMA_ALPHA,
+        pid_target_queue_depth=DYNAMIC_MEMORY_TARGET_QUEUE_DEPTH,
+        pid_penalty_factor=DYNAMIC_MEMORY_PENALTY_FACTOR,
+        pid_error_boost_factor=DYNAMIC_MEMORY_ERROR_BOOST_FACTOR,
+        rcm_memory_safety_buffer_fraction=DYNAMIC_MEMORY_RCM_MEMORY_SAFETY_BUFFER_FRACTION,
     )
 
     pipeline = RayPipeline(scaling_config=scaling_config)

{nv_ingest-2025.7.22.dev20250722 → nv_ingest-2025.7.24.dev20250724}/nv_ingest/framework/orchestration/ray/util/pipeline/stage_builders.py

@@ -57,6 +57,7 @@ from nv_ingest_api.internal.schemas.transform.transform_image_filter_schema impo
 from nv_ingest_api.internal.schemas.transform.transform_text_embedding_schema import TextEmbeddingSchema
 from nv_ingest_api.internal.schemas.transform.transform_text_splitter_schema import TextSplitterSchema
 from nv_ingest_api.util.system.hardware_info import SystemResourceProbe
+from nv_ingest.framework.orchestration.ray.util.env_config import DYNAMIC_MEMORY_THRESHOLD
 
 logger = logging.getLogger(__name__)
 
@@ -178,7 +179,7 @@ def add_pdf_extractor_stage(pipeline, default_cpu_count, stage_name="pdf_extract
     total_memory_mb = psutil.virtual_memory().total / (1024**2)
 
     # Allocate up to 75% of memory to this stage, using a 10GB high watermark per worker.
-    allocatable_memory_for_stage_mb = total_memory_mb * 0.75
+    allocatable_memory_for_stage_mb = total_memory_mb * DYNAMIC_MEMORY_THRESHOLD
     memory_based_replicas = int(allocatable_memory_for_stage_mb / 10_000.0)
 
     # Cap the number of replicas by the number of available CPU cores.
@@ -522,7 +523,7 @@ def add_text_embedding_stage(pipeline, default_cpu_count, stage_name="text_embed
         stage_actor=TextEmbeddingTransformStage,
         config=config,
         min_replicas=0,
-        max_replicas=2,
+        max_replicas=_get_max_replicas(default_cpu_count, percentage_of_cpu=0.07, replica_limit=6),
     )
 
     return stage_name
@@ -627,8 +628,22 @@ def add_source_stage(pipeline, default_cpu_count, source_name="pipeline_source")
     return source_name
 
 
-def _get_max_replicas(default_cpu_count=None, percentage_of_cpu=0.14):
+def _get_max_replicas(default_cpu_count=None, percentage_of_cpu=0.14, replica_limit=None):
+    """
+    Calculate max replicas based on CPU percentage with optional upper limit.
+
+    Args:
+        default_cpu_count (int, optional): CPU cores to use. Auto-detected if None.
+        percentage_of_cpu (float, optional): CPU percentage to allocate. Defaults to 0.14.
+        replica_limit (int, optional): Upper bound for replicas. Defaults to None.
+
+    Returns:
+        int: Maximum replicas, at least 1.
+    """
     if default_cpu_count is None:
         default_cpu_count = _system_resource_probe.get_cpu_count()
 
-    return int(max(1, (default_cpu_count * percentage_of_cpu)))
+    _max_replicas = int(max(1, (default_cpu_count * percentage_of_cpu)))
+    if replica_limit is not None:
+        _max_replicas = min(_max_replicas, replica_limit)
+    return _max_replicas

{nv_ingest-2025.7.22.dev20250722 → nv_ingest-2025.7.24.dev20250724}/nv_ingest.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 2025.7.22.dev20250722
+Version: 2025.7.24.dev20250724
 Summary: Python module for multimodal document ingestion
 Author-email: Jeremy Dyer <jdyer@nvidia.com>
 License:                                  Apache License
@@ -231,7 +231,7 @@ Requires-Dist: opentelemetry-exporter-otlp>=1.27.0
 Requires-Dist: opentelemetry-sdk>=1.27.0
 Requires-Dist: pydantic>2.0.0
 Requires-Dist: pydantic-settings>2.0.0
-Requires-Dist: pypdfium2==4.30.1
+Requires-Dist: pypdfium2==4.30.0
 Requires-Dist: pytest>=8.0.2
 Requires-Dist: pytest-mock>=3.14.0
 Requires-Dist: pytest-cov>=6.0.0
@@ -240,7 +240,6 @@ 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: ray[all]>=2.37.0
 Requires-Dist: redis>=5.2.1
 Requires-Dist: requests>=2.28.2
@@ -248,8 +247,6 @@ Requires-Dist: scikit-learn>=1.6.0
 Requires-Dist: scipy>=1.15.1
 Requires-Dist: setuptools>=78.1.1
 Requires-Dist: tabulate>=0.9.0
-Requires-Dist: torchvision
-Requires-Dist: torchaudio
 Requires-Dist: transformers>=4.47.0
 Requires-Dist: tqdm>=4.67.1
 Requires-Dist: uvicorn

{nv_ingest-2025.7.22.dev20250722 → nv_ingest-2025.7.24.dev20250724}/nv_ingest.egg-info/SOURCES.txt

@@ -71,6 +71,7 @@ nv_ingest/framework/orchestration/ray/stages/transforms/text_splitter.py
 nv_ingest/framework/orchestration/ray/stages/utility/__init__.py
 nv_ingest/framework/orchestration/ray/stages/utility/throughput_monitor.py
 nv_ingest/framework/orchestration/ray/util/__init__.py
+nv_ingest/framework/orchestration/ray/util/env_config.py
 nv_ingest/framework/orchestration/ray/util/pipeline/__init__.py
 nv_ingest/framework/orchestration/ray/util/pipeline/pid_controller.py
 nv_ingest/framework/orchestration/ray/util/pipeline/pipeline_builders.py

{nv_ingest-2025.7.22.dev20250722 → nv_ingest-2025.7.24.dev20250724}/nv_ingest.egg-info/requires.txt

@@ -16,7 +16,7 @@ opentelemetry-exporter-otlp>=1.27.0
 opentelemetry-sdk>=1.27.0
 pydantic>2.0.0
 pydantic-settings>2.0.0
-pypdfium2==4.30.1
+pypdfium2==4.30.0
 pytest>=8.0.2
 pytest-mock>=3.14.0
 pytest-cov>=6.0.0
@@ -25,7 +25,6 @@ python-docx>=1.1.2
 python-dotenv>=1.0.1
 python-pptx>=1.0.2
 prometheus-client
-torch>=2.4.1
 ray[all]>=2.37.0
 redis>=5.2.1
 requests>=2.28.2
@@ -33,8 +32,6 @@ scikit-learn>=1.6.0
 scipy>=1.15.1
 setuptools>=78.1.1
 tabulate>=0.9.0
-torchvision
-torchaudio
 transformers>=4.47.0
 tqdm>=4.67.1
 uvicorn

{nv_ingest-2025.7.22.dev20250722 → nv_ingest-2025.7.24.dev20250724}/pyproject.toml

@@ -35,7 +35,7 @@ dependencies = [
     "opentelemetry-sdk>=1.27.0",
     "pydantic>2.0.0",
     "pydantic-settings>2.0.0",
-    "pypdfium2==4.30.1",
+    "pypdfium2==4.30.0",
     "pytest>=8.0.2",
     "pytest-mock>=3.14.0",
     "pytest-cov>=6.0.0",
@@ -44,7 +44,6 @@ dependencies = [
     "python-dotenv>=1.0.1",
     "python-pptx>=1.0.2",
     "prometheus-client",
-    "torch>=2.4.1",
     "ray[all]>=2.37.0",
     "redis>=5.2.1",
     "requests>=2.28.2",
@@ -52,8 +51,6 @@ dependencies = [
     "scipy>=1.15.1",
     "setuptools>=78.1.1",
     "tabulate>=0.9.0",
-    "torchvision",
-    "torchaudio",
     "transformers>=4.47.0",
     "tqdm>=4.67.1",
     "uvicorn",