nv-ingest 25.7.7.dev20250707__py3-none-any.whl → 25.8.0rc1__py3-none-any.whl

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/framework/orchestration/ray/util/pipeline/pid_controller.py

@@ -6,9 +6,7 @@ import logging
 import math
 from dataclasses import dataclass
 
-import numpy as np
-from collections import deque
-from typing import Dict, Any, Deque, List, Tuple, Optional
+from typing import Dict, Any, List, Tuple, Optional
 
 from nv_ingest_api.util.system.hardware_info import SystemResourceProbe
 
@@ -16,7 +14,7 @@ logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
 # --- Constants ---
-DEFAULT_STAGE_COST_MB = 5000.0  # Fallback memory cost
+DEFAULT_STAGE_COST_MB = 5_000.0  # Fallback memory cost
 
 
 @dataclass
@@ -46,9 +44,7 @@ class PIDController:
         kp: float,
         ki: float,
         kd: float,  # Currently unused in delta calculation
-        stage_cost_estimates: Dict[str, int],  # Static estimates (MB)
         target_queue_depth: int = 0,
-        window_size: int = 10,
         penalty_factor: float = 0.0005,
         error_boost_factor: float = 1.5,
     ):
@@ -64,16 +60,10 @@ class PIDController:
         kd : float
             Derivative gain. Reacts to the rate of change of the error.
             (Currently set to 0 in internal calculations).
-        stage_cost_estimates : Dict[str, int]
-            Static estimated memory cost (in MB) per replica for each stage.
-            Used as a fallback and minimum for dynamic estimates.
         target_queue_depth : int, optional
             Default target queue depth for stages if not specified in metrics,
             by default 0. The PID loop tries to drive the queue depth towards
             this value.
-        window_size : int, optional
-            Number of recent samples used for dynamic memory cost estimation
-            per replica, by default 10.
         penalty_factor : float, optional
             Multiplier applied to the number of consecutive idle cycles for a
             stage. The resulting penalty effectively lowers the target queue
@@ -90,16 +80,11 @@ class PIDController:
         self.error_boost_factor = error_boost_factor
 
         # Per-Stage State
-        self.stage_cost_estimates = {
-            name: float(max(cost, 1.0)) for name, cost in stage_cost_estimates.items()  # Ensure float and min 1MB
-        }
         self.integral_error: Dict[str, float] = {}
         self.prev_error: Dict[str, float] = {}
-        self.memory_history: Dict[str, Deque[float]] = {}  # Per-replica memory history (MB)
         self.idle_cycles: Dict[str, int] = {}
 
         # Per-Stage Config
-        self.window_size = window_size
         self.penalty_factor = penalty_factor
 
     # --- Private Methods ---
@@ -110,48 +95,7 @@ class PIDController:
             logger.debug(f"[PID-{stage}] Initializing state.")
             self.integral_error[stage] = 0.0
             self.prev_error[stage] = 0.0
-            self.memory_history[stage] = deque(maxlen=self.window_size)
             self.idle_cycles[stage] = 0
-            # Ensure static cost estimate exists, provide default if missing
-            if stage not in self.stage_cost_estimates:
-                logger.warning(f"[PID-{stage}] Missing static cost estimate. Using default {DEFAULT_STAGE_COST_MB}MB.")
-                self.stage_cost_estimates[stage] = DEFAULT_STAGE_COST_MB
-
-    def _get_conservative_cost_estimate(self, stage: str) -> float:
-        """
-        Estimates dynamic memory cost, using static estimate as a floor/max.
-
-        Returns the maximum of the recent average dynamic cost per replica
-        and the static estimate provided during initialization. This provides
-        a conservative value for resource projection.
-
-        Parameters
-        ----------
-        stage : str
-            The name of the stage.
-
-        Returns
-        -------
-        float
-            The conservative memory cost estimate in MB per replica.
-        """
-        static_cost = self.stage_cost_estimates.get(stage, DEFAULT_STAGE_COST_MB)
-        memory_samples = self.memory_history.get(stage)
-
-        # Use numpy.mean if samples exist, otherwise fallback to static
-        if memory_samples and len(memory_samples) > 0:
-            try:
-                dynamic_avg = float(np.mean(memory_samples))
-                # Use max(dynamic, static) for projection, enforce min 1MB
-                cost = max(dynamic_avg, static_cost, 1.0)
-                return cost
-            except Exception as e:
-                logger.error(
-                    f"[PID-{stage}] Error calculating mean of memory samples: {e}. Falling back to static cost.",
-                    exc_info=False,
-                )
-                return max(static_cost, 1.0)  # Fallback safely
-        return max(static_cost, 1.0)  # Fallback to static estimate if no history
 
     # --- Public Method ---
 
@@ -167,8 +111,8 @@ class PIDController:
         ----------
         stage_metrics : Dict[str, Dict[str, Any]]
             Dictionary mapping stage names to their current metrics. Expected keys
-            per stage: 'replicas', 'queue_depth'. Optional: 'memory_usage',
-            'target_queue_depth', 'processing', 'min_replicas', 'max_replicas'.
+            per stage: 'replicas', 'queue_depth', 'ema_memory_per_replica'.
+            Optional: 'target_queue_depth', 'processing', 'min_replicas', 'max_replicas'.
 
         Returns
         -------
@@ -185,16 +129,9 @@ class PIDController:
 
             # --- Extract data and calculate current memory state ---
             replicas = metrics.get("replicas", 0)
-            # Start with static cost as initial guess if no memory_usage provided
-            initial_cost_guess = self.stage_cost_estimates.get(stage, DEFAULT_STAGE_COST_MB)
-            memory_usage = metrics.get("memory_usage", initial_cost_guess * max(replicas, 1))
-            # Calculate memory per replica safely (avoid division by zero)
-            current_memory_per_replica = memory_usage / max(replicas, 1.0)
-
-            # Update memory history *before* calculating the conservative cost for *this* cycle's proposal
-            self.memory_history[stage].append(current_memory_per_replica)
-            # Recalculate conservative cost *after* updating history for the proposal
-            conservative_cost = self._get_conservative_cost_estimate(stage)
+            # The conservative cost is now the EMA memory passed in from the stats collector.
+            # Fallback to a default if not present.
+            conservative_cost = metrics.get("ema_memory_per_replica", DEFAULT_STAGE_COST_MB)
 
             # --- PID Calculation ---
             queue_depth = metrics.get("queue_depth", 0)
@@ -296,7 +233,6 @@ class ResourceConstraintManager:
         self,
         max_replicas: int,
         memory_threshold: int,
-        estimated_edge_cost_mb: int,
         memory_safety_buffer_fraction: float,
     ):
         """
@@ -309,7 +245,6 @@ class ResourceConstraintManager:
 
         self.max_replicas = max_replicas
         self.memory_threshold_mb = memory_threshold
-        self.estimated_edge_cost_mb = estimated_edge_cost_mb  # Keep track, though unused
         self.memory_safety_buffer_fraction = memory_safety_buffer_fraction  # Unused
         self.effective_memory_limit_mb = self.memory_threshold_mb
 

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.
@@ -78,15 +81,17 @@ class PipelineCreationSchema(BaseModel):
     otel_exporter_otlp_endpoint: str = os.getenv("OTEL_EXPORTER_OTLP_ENDPOINT", "localhost:4317")
 
     # OCR settings
-    paddle_http_endpoint: str = os.getenv("PADDLE_HTTP_ENDPOINT", "https://ai.api.nvidia.com/v1/cv/baidu/paddleocr")
-    paddle_infer_protocol: str = os.getenv("PADDLE_INFER_PROTOCOL", "http")
+    ocr_http_endpoint: str = os.getenv("OCR_HTTP_ENDPOINT", "https://ai.api.nvidia.com/v1/cv/baidu/paddleocr")
+    ocr_infer_protocol: str = os.getenv("OCR_INFER_PROTOCOL", "http")
+    ocr_model_name: str = os.getenv("OCR_MODEL_NAME", "paddle")
 
     # Task queue settings
     REDIS_INGEST_TASK_QUEUE: str = "ingest_task_queue"
 
     # 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"
+        "VLM_CAPTION_ENDPOINT",
+        "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")
 
@@ -233,7 +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/framework/orchestration/ray/util/pipeline/stage_builders.py

@@ -3,7 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import os
-
+import psutil
 import click
 import logging
 
@@ -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__)
 
@@ -174,6 +175,16 @@ def add_metadata_injector_stage(pipeline, default_cpu_count, stage_name="metadat
 
 
 def add_pdf_extractor_stage(pipeline, default_cpu_count, stage_name="pdf_extractor"):
+    # Heuristic: Determine max_replicas based on system memory, capped by CPU cores.
+    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 * 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.
+    max_replicas = max(1, min(memory_based_replicas, default_cpu_count))
+
     yolox_grpc, yolox_http, yolox_auth, yolox_protocol = get_nim_service("yolox")
     nemoretriever_parse_grpc, nemoretriever_parse_http, nemoretriever_parse_auth, nemoretriever_parse_protocol = (
         get_nim_service("nemoretriever_parse")
@@ -203,9 +214,8 @@ def add_pdf_extractor_stage(pipeline, default_cpu_count, stage_name="pdf_extract
         stage_actor=PDFExtractorStage,
         config=extractor_config,
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 3))),  # 33% of available CPU cores
+        max_replicas=max_replicas,
     )
-
     return stage_name
 
 
@@ -213,15 +223,15 @@ def add_table_extractor_stage(pipeline, default_cpu_count, stage_name="table_ext
     yolox_table_structure_grpc, yolox_table_structure_http, yolox_auth, yolox_table_structure_protocol = (
         get_nim_service("yolox_table_structure")
     )
-    paddle_grpc, paddle_http, paddle_auth, paddle_protocol = get_nim_service("paddle")
+    ocr_grpc, ocr_http, ocr_auth, ocr_protocol = get_nim_service("ocr")
 
     table_extractor_config = TableExtractorSchema(
         **{
             "endpoint_config": {
                 "yolox_endpoints": (yolox_table_structure_grpc, yolox_table_structure_http),
                 "yolox_infer_protocol": yolox_table_structure_protocol,
-                "paddle_endpoints": (paddle_grpc, paddle_http),
-                "paddle_infer_protocol": paddle_protocol,
+                "ocr_endpoints": (ocr_grpc, ocr_http),
+                "ocr_infer_protocol": ocr_protocol,
                 "auth_token": yolox_auth,
             }
         }
@@ -232,7 +242,7 @@ def add_table_extractor_stage(pipeline, default_cpu_count, stage_name="table_ext
         stage_actor=TableExtractorStage,
         config=table_extractor_config,
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 7))),  # 14% of available CPU cores
+        max_replicas=_get_max_replicas(default_cpu_count, percentage_of_cpu=0.20),
     )
 
     return stage_name
@@ -242,15 +252,15 @@ def add_chart_extractor_stage(pipeline, default_cpu_count, stage_name="chart_ext
     yolox_graphic_elements_grpc, yolox_graphic_elements_http, yolox_auth, yolox_graphic_elements_protocol = (
         get_nim_service("yolox_graphic_elements")
     )
-    paddle_grpc, paddle_http, paddle_auth, paddle_protocol = get_nim_service("paddle")
+    ocr_grpc, ocr_http, ocr_auth, ocr_protocol = get_nim_service("ocr")
 
     chart_extractor_config = ChartExtractorSchema(
         **{
             "endpoint_config": {
                 "yolox_endpoints": (yolox_graphic_elements_grpc, yolox_graphic_elements_http),
                 "yolox_infer_protocol": yolox_graphic_elements_protocol,
-                "paddle_endpoints": (paddle_grpc, paddle_http),
-                "paddle_infer_protocol": paddle_protocol,
+                "ocr_endpoints": (ocr_grpc, ocr_http),
+                "ocr_infer_protocol": ocr_protocol,
                 "auth_token": yolox_auth,
             }
         }
@@ -261,21 +271,21 @@ def add_chart_extractor_stage(pipeline, default_cpu_count, stage_name="chart_ext
         stage_actor=ChartExtractorStage,
         config=chart_extractor_config,
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 7))),  # 14% of available CPU cores
+        max_replicas=_get_max_replicas(default_cpu_count, percentage_of_cpu=0.20),
     )
 
     return stage_name
 
 
 def add_infographic_extractor_stage(pipeline, default_cpu_count, stage_name="infographic_extractor"):
-    paddle_grpc, paddle_http, paddle_auth, paddle_protocol = get_nim_service("paddle")
+    ocr_grpc, ocr_http, ocr_auth, ocr_protocol = get_nim_service("ocr")
 
     infographic_content_extractor_config = InfographicExtractorSchema(
         **{
             "endpoint_config": {
-                "paddle_endpoints": (paddle_grpc, paddle_http),
-                "paddle_infer_protocol": paddle_protocol,
-                "auth_token": paddle_auth,
+                "ocr_endpoints": (ocr_grpc, ocr_http),
+                "ocr_infer_protocol": ocr_protocol,
+                "auth_token": ocr_auth,
             }
         }
     )
@@ -285,7 +295,7 @@ def add_infographic_extractor_stage(pipeline, default_cpu_count, stage_name="inf
         stage_actor=InfographicExtractorStage,
         config=infographic_content_extractor_config,
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 14))),  # 7% of available CPU cores
+        max_replicas=2,
     )
 
     return stage_name
@@ -307,7 +317,7 @@ def add_image_extractor_stage(pipeline, default_cpu_count, stage_name="image_ext
         stage_actor=ImageExtractorStage,
         config=image_extractor_config,
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 14))),  # 7% of available CPU cores
+        max_replicas=2,
     )
 
     return stage_name
@@ -329,7 +339,7 @@ def add_docx_extractor_stage(pipeline, default_cpu_count, stage_name="docx_extra
         stage_actor=DocxExtractorStage,
         config=DocxExtractorSchema(**docx_extractor_config),
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 14))),  # 7% of available CPU cores
+        max_replicas=2,
     )
 
     return stage_name
@@ -351,7 +361,7 @@ def add_pptx_extractor_stage(pipeline, default_cpu_count, stage_name="pptx_extra
         stage_actor=PPTXExtractorStage,
         config=PPTXExtractorSchema(**pptx_extractor_config),
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 14))),  # 7% of available CPU cores
+        max_replicas=2,
     )
 
     return stage_name
@@ -373,11 +383,7 @@ def add_audio_extractor_stage(pipeline, default_cpu_count, stage_name="audio_ext
     )
 
     pipeline.add_stage(
-        name=stage_name,
-        stage_actor=AudioExtractorStage,
-        config=audio_extractor_config,
-        min_replicas=0,
-        max_replicas=1,  # Audio extraction is a heavy IO bound operation with minimal CPU usage
+        name=stage_name, stage_actor=AudioExtractorStage, config=audio_extractor_config, min_replicas=0, max_replicas=2
     )
 
     return stage_name
@@ -390,7 +396,7 @@ def add_html_extractor_stage(pipeline, default_cpu_count, stage_name="html_extra
         stage_actor=HtmlExtractorStage,
         config=HtmlExtractorSchema(),
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 14))),  # 7% of available CPU cores
+        max_replicas=2,
     )
 
     return stage_name
@@ -455,7 +461,7 @@ def add_text_splitter_stage(pipeline, default_cpu_count, stage_name="text_splitt
         stage_actor=TextSplitterStage,
         config=config,
         min_replicas=0,
-        max_replicas=int(max(1, (default_cpu_count // 14))),  # 7% of available CPU cores
+        max_replicas=2,
     )
 
     return stage_name
@@ -517,7 +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=int(max(1, (default_cpu_count // 14))),  # 7% of available CPU cores
+        max_replicas=_get_max_replicas(default_cpu_count, percentage_of_cpu=0.07, replica_limit=6),
     )
 
     return stage_name
@@ -620,3 +626,24 @@ def add_source_stage(pipeline, default_cpu_count, source_name="pipeline_source")
         start_simple_message_broker(source_config.broker_client.model_dump())
 
     return source_name
+
+
+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()
+
+    _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-25.7.7.dev20250707.dist-info → nv_ingest-25.8.0rc1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nv-ingest
-Version: 25.7.7.dev20250707
+Version: 25.8.0rc1
 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-25.7.7.dev20250707.dist-info → nv_ingest-25.8.0rc1.dist-info}/RECORD

@@ -3,7 +3,7 @@ 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
-nv_ingest/api/v1/health.py,sha256=zqu-isMRjh4NveS4XWh5FaAZGPIlBVxpCOg3Uu8nUHQ,4746
+nv_ingest/api/v1/health.py,sha256=pV-RoVq5y0iBPp0qZoLzd1xKpd0JiHAi0UMyMj99LqU,4740
 nv_ingest/api/v1/ingest.py,sha256=LWk3LN4lBd3uO8h30EN42g3LHCVcO00avVd5ohVK7NI,19392
 nv_ingest/api/v1/metrics.py,sha256=ZGVRApYLnzc2f2C7wRgGd7deqiXan-jxfA-33a16clY,981
 nv_ingest/framework/__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=qUNvWiNBUEEzuCySY3usWmHQz9qMgTGVZuKmLWqTsi4,16412
+nv_ingest/framework/orchestration/ray/examples/pipeline_test_harness.py,sha256=DufjmNm-05uTkq_Mz0QQB6fHw_Rl9eX3PRtnH4sntGs,16405
 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=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/primitives/pipeline_topology.py,sha256=IxLQYHYc9BnMOi73TSJzuhl8KOJAwlwwp8SPdUkV3nE,15737
+nv_ingest/framework/orchestration/ray/primitives/ray_pipeline.py,sha256=HGJ_TyLTKKRl10HWfyx3D-n-zrFY0Fg9TN74UbOeCm8,66584
+nv_ingest/framework/orchestration/ray/primitives/ray_stat_collector.py,sha256=8SpZzulHatqah7U3YHJMTLaYyPlWdCoaer_oNjhmHZo,17221
 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
@@ -40,7 +40,7 @@ nv_ingest/framework/orchestration/ray/stages/meta/__init__.py,sha256=wQSlVx3T14Z
 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=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/meta/ray_actor_stage_base.py,sha256=pvBFsURWoDiAmDWNTLv2pdm5slv-1OnuXxwYvgaKumU,25703
 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
 nv_ingest/framework/orchestration/ray/stages/mutate/image_filter.py,sha256=9ek5rVa4_GVdmVHGMJvbxacRSpIqVoUxgv28lzJwrTQ,3319
@@ -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=9YoVytbFFt-RpIR_MN2m3T93zVTjts8tjhi0qzLJkTw,19922
+nv_ingest/framework/orchestration/ray/stages/sources/message_broker_task_source.py,sha256=b9ndnQBB1paR0iRe3NdzQ7BZ2S65LG2jbtjXvvDc_s4,21183
 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
@@ -63,11 +63,12 @@ nv_ingest/framework/orchestration/ray/stages/transforms/text_splitter.py,sha256=
 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/env_config.py,sha256=GN9msJ_3jdOBIAPnXNxX0ds_BKtHRnRhnYxwzcAU2KY,2386
 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/pid_controller.py,sha256=flRLS7yc5n6gheykayuL3prC7O-ZhcVY2s9Wc14SGWE,47377
 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/pipeline_runners.py,sha256=-lXiAjxpk3CxIFWi7GW8ggTn1cjyugYZ9EtSfTT20mg,14844
+nv_ingest/framework/orchestration/ray/util/pipeline/stage_builders.py,sha256=Vgwcb4X-k-PyQKnRMjLxWehhWwE67G8w7Q-1h7EPoL0,22212
 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
@@ -96,8 +97,8 @@ 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-25.7.7.dev20250707.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
-nv_ingest-25.7.7.dev20250707.dist-info/METADATA,sha256=6cvQcvjAZv_9WRvsAHUMu_tRKsSI2LLfgBKY5eCioT0,15139
-nv_ingest-25.7.7.dev20250707.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-nv_ingest-25.7.7.dev20250707.dist-info/top_level.txt,sha256=sjb0ajIsgn3YgftSjZHlYO0HjYAIIhNuXG_AmywCvaU,10
-nv_ingest-25.7.7.dev20250707.dist-info/RECORD,,
+nv_ingest-25.8.0rc1.dist-info/licenses/LICENSE,sha256=xx0jnfkXJvxRnG63LTGOxlggYnIysveWIZ6H3PNdCrQ,11357
+nv_ingest-25.8.0rc1.dist-info/METADATA,sha256=6SNNz3Z9ZytUfWSmUO4pSqboflJssA_CACRAUM5dLSQ,15049
+nv_ingest-25.8.0rc1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+nv_ingest-25.8.0rc1.dist-info/top_level.txt,sha256=sjb0ajIsgn3YgftSjZHlYO0HjYAIIhNuXG_AmywCvaU,10
+nv_ingest-25.8.0rc1.dist-info/RECORD,,