openlit 1.13.0__tar.gz → 1.14.0__tar.gz

{openlit-1.13.0 → openlit-1.14.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: openlit
-Version: 1.13.0
+Version: 1.14.0
 Summary: OpenTelemetry-native Auto instrumentation library for monitoring LLM Applications, facilitating the integration of observability into your GenAI-driven projects
 Home-page: https://github.com/openlit/openlit/tree/main/openlit/python
 Keywords: OpenTelemetry,otel,otlp,llm,tracing,openai,anthropic,claude,cohere,llm monitoring,observability,monitoring,gpt,Generative AI,chatGPT
@@ -14,11 +14,13 @@ Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Requires-Dist: boto3 (>=1.34.0,<2.0.0)
 Requires-Dist: botocore (>=1.34.0,<2.0.0)
+Requires-Dist: gpustat (>=1.1.1,<2.0.0)
 Requires-Dist: opentelemetry-api (>=1.24.0,<2.0.0)
 Requires-Dist: opentelemetry-exporter-otlp (>=1.24.0,<2.0.0)
 Requires-Dist: opentelemetry-instrumentation (>=0.45b0,<0.46)
 Requires-Dist: opentelemetry-sdk (>=1.24.0,<2.0.0)
 Requires-Dist: requests (>=2.26.0,<3.0.0)
+Requires-Dist: schedule (>=1.2.2,<2.0.0)
 Requires-Dist: tiktoken (>=0.6.0,<0.7.0)
 Project-URL: Repository, https://github.com/openlit/openlit/tree/main/openlit/python
 Description-Content-Type: text/markdown
@@ -185,6 +187,7 @@ Below is a detailed overview of the configuration options available, allowing yo
 | `disabled_instrumentors`| List of instrumentors to disable. | `None` |    No    |
 | `disable_metrics`       | If set, disables the collection of metrics.                                                   | `False`        |    No    |
 | `pricing_json`          | URL or file path of the pricing JSON file.                                             | `https://github.com/openlit/openlit/blob/main/assets/pricing.json`        |    No    |
+| `collect_gpu_stats`          | Flag to enable or disable GPU metrics collection.                                         | `False`        |    No    |
 
 ## 🌱 Contributing
 

{openlit-1.13.0 → openlit-1.14.0}/README.md

@@ -160,6 +160,7 @@ Below is a detailed overview of the configuration options available, allowing yo
 | `disabled_instrumentors`| List of instrumentors to disable. | `None` |    No    |
 | `disable_metrics`       | If set, disables the collection of metrics.                                                   | `False`        |    No    |
 | `pricing_json`          | URL or file path of the pricing JSON file.                                             | `https://github.com/openlit/openlit/blob/main/assets/pricing.json`        |    No    |
+| `collect_gpu_stats`          | Flag to enable or disable GPU metrics collection.                                         | `False`        |    No    |
 
 ## 🌱 Contributing
 

{openlit-1.13.0 → openlit-1.14.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "openlit"
-version = "1.13.0"
+version = "1.14.0"
 description = "OpenTelemetry-native Auto instrumentation library for monitoring LLM Applications, facilitating the integration of observability into your GenAI-driven projects"
 authors = ["OpenLIT"]
 repository = "https://github.com/openlit/openlit/tree/main/openlit/python"
@@ -11,7 +11,9 @@ keywords = ["OpenTelemetry", "otel", "otlp","llm", "tracing", "openai", "anthrop
 [tool.poetry.dependencies]
 python = "^3.7.1"
 requests = "^2.26.0"
+schedule = "^1.2.2"
 tiktoken = "^0.6.0"
+gpustat = "^1.1.1"
 boto3 = "^1.34.0"
 botocore = "^1.34.0"
 opentelemetry-api = "^1.24.0"

{openlit-1.13.0 → openlit-1.14.0}/src/openlit/__init__.py

@@ -40,6 +40,7 @@ from openlit.instrumentation.pinecone import PineconeInstrumentor
 from openlit.instrumentation.qdrant import QdrantInstrumentor
 from openlit.instrumentation.milvus import MilvusInstrumentor
 from openlit.instrumentation.transformers import TransformersInstrumentor
+from openlit.instrumentation.gpu import NvidiaGPUInstrumentor
 
 # Set up logging for error and information messages.
 logger = logging.getLogger(__name__)
@@ -155,20 +156,9 @@ def instrument_if_available(
         except Exception as e:
             logger.error("Failed to instrument %s: %s", instrumentor_name, e)
 
-
-def init(
-    environment="default",
-    application_name="default",
-    tracer=None,
-    otlp_endpoint=None,
-    otlp_headers=None,
-    disable_batch=False,
-    trace_content=True,
-    disabled_instrumentors=None,
-    meter=None,
-    disable_metrics=False,
-    pricing_json=None,
-):
+def init(environment="default", application_name="default", tracer=None, otlp_endpoint=None,
+         otlp_headers=None, disable_batch=False, trace_content=True, disabled_instrumentors=None,
+         meter=None, disable_metrics=False, pricing_json=None, collect_gpu_stats=False):
     """
     Initializes the openLIT configuration and setups tracing.
 
@@ -185,8 +175,9 @@ def init(
         disable_batch (bool): Flag to disable batch span processing (Optional).
         trace_content (bool): Flag to trace content (Optional).
         disabled_instrumentors (List[str]): Optional. List of instrumentor names to disable.
-        disable_metrics (bool): Flag to disable metrics (Optional)
-        pricing_json(str): File path or url to the pricing json (Optional)
+        disable_metrics (bool): Flag to disable metrics (Optional).
+        pricing_json(str): File path or url to the pricing json (Optional).
+        collect_gpu_stats (bool): Flag to enable or disable GPU metrics collection.
     """
     disabled_instrumentors = disabled_instrumentors if disabled_instrumentors else []
     # Check for invalid instrumentor names
@@ -289,8 +280,13 @@ def init(
 
         # Initialize and instrument only the enabled instrumentors
         for name, instrumentor in instrumentor_instances.items():
-            instrument_if_available(
-                name, instrumentor, config, disabled_instrumentors, module_name_map
+            instrument_if_available(name, instrumentor, config,
+                                    disabled_instrumentors, module_name_map)
+
+        if (disable_metrics is False) and (collect_gpu_stats is True):
+            NvidiaGPUInstrumentor().instrument(
+                environment=config.environment,
+                application_name=config.application_name,
             )
 
     except Exception as e:

openlit-1.14.0/src/openlit/instrumentation/gpu/__init__.py

@@ -0,0 +1,132 @@
+# pylint: disable=useless-return, bad-staticmethod-argument, duplicate-code, import-outside-toplevel, broad-exception-caught, unused-argument
+"""Initializer of Auto Instrumentation of GPU Metrics"""
+
+from typing import Collection, Iterable
+import logging
+from functools import partial
+
+from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
+from opentelemetry.sdk.resources import TELEMETRY_SDK_NAME
+from opentelemetry.metrics import get_meter, CallbackOptions, Observation
+
+from openlit.semcov import SemanticConvetion
+
+# Initialize logger for logging potential issues and operations
+logger = logging.getLogger(__name__)
+
+class NvidiaGPUInstrumentor(BaseInstrumentor):
+    """
+    An instrumentor for collecting NVIDIA GPU metrics.
+    """
+
+    def instrumentation_dependencies(self) -> Collection[str]:
+        return []
+
+    def _instrument(self, **kwargs):
+
+        application_name = kwargs.get("application_name", "default")
+        environment = kwargs.get("environment", "default")
+
+        meter = get_meter(
+            __name__,
+            "0.1.0",
+            schema_url="https://opentelemetry.io/schemas/1.11.0",
+        )
+
+        def check_and_record(value):
+            return value if value is not None else 0
+
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_UTILIZATION,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "utilization")],
+            description="GPU Utilization",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_UTILIZATION_ENC,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "utilization_enc")],
+            description="GPU Encoder Utilization",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_UTILIZATION_DEC,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "utilization_dec")],
+            description="GPU Decoder Utilization",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_TEMPERATURE,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "temperature")],
+            description="GPU Temperature",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_FAN_SPEED,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "fan_speed")],
+            description="GPU Fan Speed",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_MEMORY_AVAILABLE,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "memory_available")],
+            description="GPU Memory Available",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_MEMORY_TOTAL,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "memory_total")],
+            description="GPU Memory Total",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_MEMORY_USED,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "memory_used")],
+            description="GPU Memory Used",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_MEMORY_FREE,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "memory_free")],
+            description="GPU Memory Free",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_POWER_DRAW,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "power_draw")],
+            description="GPU Power Draw",
+        )
+        meter.create_observable_gauge(
+            name=SemanticConvetion.GPU_POWER_LIMIT,
+            callbacks=[partial(self._collect_metric, environment,
+                               application_name, check_and_record, "power_limit")],
+            description="GPU Power Limit",
+        )
+
+    def _uninstrument(self, **kwargs):
+        # Proper uninstrumentation logic to revert patched methods
+        pass
+
+    def _collect_metric(self, environment, application_name,
+                        check_and_record, metric_name,
+                        options: CallbackOptions) -> Iterable[Observation]:
+
+        import gpustat
+
+        try:
+            gpu_stats = gpustat.GPUStatCollection.new_query()
+
+            for gpu in gpu_stats.gpus:
+                attributes = {
+                    TELEMETRY_SDK_NAME: "openlit",
+                    SemanticConvetion.GEN_AI_APPLICATION_NAME: application_name,
+                    SemanticConvetion.GEN_AI_ENVIRONMENT: environment,
+                    SemanticConvetion.GPU_INDEX: gpu.index,
+                    SemanticConvetion.GPU_UUID: gpu.uuid,
+                    SemanticConvetion.GPU_NAME: gpu.name,
+                }
+
+                yield Observation(check_and_record(getattr(gpu, metric_name, 0)), attributes)
+
+        except Exception as e:
+            logger.error("Error in GPU metrics collection: %s", e)

{openlit-1.13.0 → openlit-1.14.0}/src/openlit/semcov/__init__.py

@@ -145,3 +145,20 @@ class SemanticConvetion:
     DB_SYSTEM_PINECONE = "pinecone"
     DB_SYSTEM_QDRANT = "qdrant"
     DB_SYSTEM_MILVUS = "milvus"
+
+    # GPU
+    GPU_INDEX = "gpu.index"
+    GPU_UUID = "gpu.uuid"
+    GPU_NAME = "gpu.name"
+
+    GPU_UTILIZATION = "gpu.utilization"
+    GPU_UTILIZATION_ENC = "gpu.enc.utilization"
+    GPU_UTILIZATION_DEC = "gpu.dec.utilization"
+    GPU_TEMPERATURE = "gpu.temperature"
+    GPU_FAN_SPEED = "gpu.fan_speed"
+    GPU_MEMORY_AVAILABLE = "gpu.memory.available"
+    GPU_MEMORY_TOTAL = "gpu.memory.total"
+    GPU_MEMORY_USED = "gpu.memory.used"
+    GPU_MEMORY_FREE = "gpu.memory.free"
+    GPU_POWER_DRAW = "gpu.power.draw"
+    GPU_POWER_LIMIT = "gpu.power.limit"