openlit 1.34.12__tar.gz → 1.34.14__tar.gz

{openlit-1.34.12 → openlit-1.34.14}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: openlit
-Version: 1.34.12
+Version: 1.34.14
 Summary: OpenTelemetry-native Auto instrumentation library for monitoring LLM Applications and GPUs, facilitating the integration of observability into your GenAI-driven projects
 License: Apache-2.0
 Keywords: OpenTelemetry,otel,otlp,llm,tracing,openai,anthropic,claude,cohere,llm monitoring,observability,monitoring,gpt,Generative AI,chatGPT,gpu

{openlit-1.34.12 → openlit-1.34.14}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "openlit"
-version = "1.34.12"
+version = "1.34.14"
 description = "OpenTelemetry-native Auto instrumentation library for monitoring LLM Applications and GPUs, facilitating the integration of observability into your GenAI-driven projects"
 authors = ["OpenLIT"]
 license = "Apache-2.0"

{openlit-1.34.12 → openlit-1.34.14}/src/openlit/instrumentation/transformers/__init__.py

@@ -30,12 +30,19 @@ class TransformersInstrumentor(BaseInstrumentor):
         version = importlib.metadata.version("transformers")
 
         wrap_function_wrapper(
-            "transformers",  
-            "TextGenerationPipeline.__call__",  
-            pipeline_wrapper(version, environment, application_name,
-                 tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+            "transformers",
+            "TextGenerationPipeline.__call__",
+            pipeline_wrapper(
+                version,
+                environment,
+                application_name,
+                tracer,
+                pricing_info,
+                capture_message_content,
+                metrics,
+                disable_metrics
+            ),
         )
 
     def _uninstrument(self, **kwargs):
-        # Proper uninstrumentation logic to revert patched methods
         pass

openlit-1.34.14/src/openlit/instrumentation/transformers/transformers.py

@@ -0,0 +1,53 @@
+"""
+Module for monitoring HF Transformers API calls.
+"""
+
+import time
+from opentelemetry.trace import SpanKind
+from openlit.__helpers import set_server_address_and_port
+from openlit.instrumentation.transformers.utils import process_chat_response
+from openlit.semcov import SemanticConvention
+
+
+def pipeline_wrapper(version, environment, application_name, tracer, pricing_info,
+                     capture_message_content, metrics, disable_metrics):
+    """
+    Generates a telemetry wrapper for GenAI function call
+    """
+
+    def wrapper(wrapped, instance, args, kwargs):
+        """
+        Wraps the GenAI function call.
+        """
+
+        server_address, server_port = set_server_address_and_port(instance, "127.0.0.1", 80)
+        request_model = instance.model.config.name_or_path
+
+        span_name = f"{SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT} {request_model}"
+
+        with tracer.start_as_current_span(span_name, kind=SpanKind.CLIENT) as span:
+            start_time = time.time()
+            response = wrapped(*args, **kwargs)
+
+            response = process_chat_response(
+                instance=instance,
+                response=response,
+                request_model=request_model,
+                pricing_info=pricing_info,
+                server_port=server_port,
+                server_address=server_address,
+                environment=environment,
+                application_name=application_name,
+                metrics=metrics,
+                start_time=start_time,
+                span=span,
+                args=args,
+                kwargs=kwargs,
+                capture_message_content=capture_message_content,
+                disable_metrics=disable_metrics,
+                version=version,
+            )
+
+        return response
+
+    return wrapper

openlit-1.34.14/src/openlit/instrumentation/transformers/utils.py

@@ -0,0 +1,199 @@
+"""
+HF Transformers OpenTelemetry instrumentation utility functions
+"""
+import time
+
+from opentelemetry.trace import Status, StatusCode
+
+from openlit.__helpers import (
+    general_tokens,
+    get_chat_model_cost,
+    common_span_attributes,
+    record_completion_metrics,
+)
+from openlit.semcov import SemanticConvention
+
+def format_content(content):
+    """
+    Format content to a consistent structure.
+    """
+    if isinstance(content, str):
+        return content
+    elif isinstance(content, list):
+        # Check if its a list of chat messages (like in the test case)
+        if (len(content) > 0 and isinstance(content[0], dict) and
+            "role" in content[0] and "content" in content[0]):
+            # Handle chat message format like Groq
+            formatted_messages = []
+            for message in content:
+                role = message["role"]
+                msg_content = message["content"]
+
+                if isinstance(msg_content, list):
+                    content_str = ", ".join(
+                        f'{item["type"]}: {item["text"] if "text" in item else item.get("image_url", str(item))}'
+                        if isinstance(item, dict) and "type" in item
+                        else str(item)
+                        for item in msg_content
+                    )
+                    formatted_messages.append(f"{role}: {content_str}")
+                else:
+                    formatted_messages.append(f"{role}: {msg_content}")
+            return "\n".join(formatted_messages)
+        else:
+            # Handle other list formats (transformers responses)
+            formatted_content = []
+            for item in content:
+                if isinstance(item, str):
+                    formatted_content.append(item)
+                elif isinstance(item, dict):
+                    # Handle dict format for transformers
+                    if "generated_text" in item:
+                        formatted_content.append(str(item["generated_text"]))
+                    else:
+                        formatted_content.append(str(item))
+                else:
+                    formatted_content.append(str(item))
+            return " ".join(formatted_content)
+    else:
+        return str(content)
+
+def common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+    capture_message_content, disable_metrics, version, args, kwargs, is_stream):
+
+    """
+    Process chat request and generate Telemetry
+    """
+
+    scope._end_time = time.time()
+    forward_params = scope._instance._forward_params
+    request_model = scope._instance.model.config.name_or_path
+
+    input_tokens = general_tokens(scope._prompt)
+    output_tokens = general_tokens(scope._completion)
+
+    cost = get_chat_model_cost(request_model, pricing_info, input_tokens, output_tokens)
+
+    # Common Span Attributes
+    common_span_attributes(scope,
+        SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT, SemanticConvention.GEN_AI_SYSTEM_HUGGING_FACE,
+        scope._server_address, scope._server_port, request_model, request_model,
+        environment, application_name, is_stream, scope._tbt, scope._ttft, version)
+
+    # Set request parameters from forward_params
+    if forward_params.get("temperature") is not None:
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, forward_params["temperature"])
+    if forward_params.get("top_k") is not None:
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_TOP_K, forward_params["top_k"])
+    if forward_params.get("top_p") is not None:
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_TOP_P, forward_params["top_p"])
+    if forward_params.get("max_length") is not None:
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, forward_params["max_length"])
+
+    # Set token usage and cost attributes
+    scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_INPUT_TOKENS, input_tokens)
+    scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_OUTPUT_TOKENS, output_tokens)
+    scope._span.set_attribute(SemanticConvention.GEN_AI_CLIENT_TOKEN_USAGE, input_tokens + output_tokens)
+    scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_COST, cost)
+
+    # Span Attributes for Content
+    if capture_message_content:
+        scope._span.set_attribute(SemanticConvention.GEN_AI_CONTENT_PROMPT, scope._prompt)
+        scope._span.set_attribute(SemanticConvention.GEN_AI_CONTENT_COMPLETION, scope._completion)
+
+        # To be removed once the change to span_attributes (from span events) is complete
+        scope._span.add_event(
+            name=SemanticConvention.GEN_AI_CONTENT_PROMPT_EVENT,
+            attributes={
+                SemanticConvention.GEN_AI_CONTENT_PROMPT: scope._prompt,
+            },
+        )
+        scope._span.add_event(
+            name=SemanticConvention.GEN_AI_CONTENT_COMPLETION_EVENT,
+            attributes={
+                SemanticConvention.GEN_AI_CONTENT_COMPLETION: scope._completion,
+            },
+        )
+
+    scope._span.set_status(Status(StatusCode.OK))
+
+    # Record metrics using the standardized helper function
+    if not disable_metrics:
+        record_completion_metrics(metrics, SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT,
+            SemanticConvention.GEN_AI_SYSTEM_HUGGING_FACE, scope._server_address, scope._server_port,
+            request_model, request_model, environment, application_name, scope._start_time, scope._end_time,
+            cost, input_tokens, output_tokens, scope._tbt, scope._ttft)
+
+def process_chat_response(instance, response, request_model, pricing_info, server_port, server_address,
+    environment, application_name, metrics, start_time,
+    span, args, kwargs, capture_message_content=False, disable_metrics=False, version="1.0.0"):
+    """
+    Process chat request and generate Telemetry
+    """
+
+    scope = type("GenericScope", (), {})()
+    scope._instance = instance
+    scope._start_time = start_time
+    scope._end_time = time.time()
+    scope._span = span
+    scope._server_address = server_address
+    scope._server_port = server_port
+    scope._kwargs = kwargs
+    scope._args = args
+
+    # Extract prompt from args or kwargs
+    if args and len(args) > 0:
+        scope._prompt = args[0]
+    else:
+        scope._prompt = (
+            kwargs.get("text_inputs") or
+            (kwargs.get("image") and kwargs.get("question") and
+             ("image: " + kwargs.get("image") + " question:" + kwargs.get("question"))) or
+            kwargs.get("fallback") or
+            ""
+        )
+    scope._prompt = format_content(scope._prompt)
+
+    # Process response based on task type
+    task = kwargs.get("task", "text-generation")
+
+    if task == "text-generation":
+        # Handle text generation responses
+        if isinstance(response, list) and len(response) > 0:
+            first_entry = response[0]
+            if isinstance(first_entry, dict):
+                if isinstance(first_entry.get("generated_text"), list):
+                    # Handle nested list format
+                    last_element = first_entry.get("generated_text")[-1]
+                    scope._completion = last_element.get("content", str(last_element))
+                else:
+                    # Handle standard format
+                    scope._completion = first_entry.get("generated_text", "")
+            else:
+                scope._completion = str(first_entry)
+        else:
+            scope._completion = ""
+
+    elif task == "automatic-speech-recognition":
+        scope._completion = response.get("text", "") if isinstance(response, dict) else ""
+
+    elif task == "image-classification":
+        scope._completion = str(response[0]) if isinstance(response, list) and len(response) > 0 else ""
+
+    elif task == "visual-question-answering":
+        if isinstance(response, list) and len(response) > 0 and isinstance(response[0], dict):
+            scope._completion = response[0].get("answer", "")
+        else:
+            scope._completion = ""
+    else:
+        # Default handling for other tasks
+        scope._completion = format_content(response)
+
+    # Initialize timing attributes
+    scope._tbt = 0
+    scope._ttft = scope._end_time - scope._start_time
+
+    common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+                     capture_message_content, disable_metrics, version, args, kwargs, is_stream=False)
+
+    return response

{openlit-1.34.12 → openlit-1.34.14}/src/openlit/instrumentation/vllm/__init__.py

@@ -1,4 +1,3 @@
-# pylint: disable=useless-return, bad-staticmethod-argument, disable=duplicate-code
 """Initializer of Auto Instrumentation of vLLM Functions"""
 
 from typing import Collection
@@ -14,15 +13,15 @@ _instruments = ("vllm >= 0.5.4",)
 
 class VLLMInstrumentor(BaseInstrumentor):
     """
-    An instrumentor for vLLM's client library.
+    An instrumentor for vLLM client library.
     """
 
     def instrumentation_dependencies(self) -> Collection[str]:
         return _instruments
 
     def _instrument(self, **kwargs):
-        application_name = kwargs.get("application_name", "default_application")
-        environment = kwargs.get("environment", "default_environment")
+        application_name = kwargs.get("application_name", "default")
+        environment = kwargs.get("environment", "default")
         tracer = kwargs.get("tracer")
         metrics = kwargs.get("metrics_dict")
         pricing_info = kwargs.get("pricing_info", {})
@@ -30,14 +29,13 @@ class VLLMInstrumentor(BaseInstrumentor):
         disable_metrics = kwargs.get("disable_metrics")
         version = importlib.metadata.version("vllm")
 
-        # sync chat
+        # Chat completions
         wrap_function_wrapper(
             "vllm.entrypoints.llm",
             "LLM.generate",
             generate(version, environment, application_name,
-                  tracer, pricing_info, capture_message_content, metrics, disable_metrics),
+                     tracer, pricing_info, capture_message_content, metrics, disable_metrics),
         )
 
     def _uninstrument(self, **kwargs):
-        # Proper uninstrumentation logic to revert patched methods
         pass

openlit-1.34.14/src/openlit/instrumentation/vllm/utils.py

@@ -0,0 +1,143 @@
+"""
+vLLM OpenTelemetry instrumentation utility functions
+"""
+import time
+
+from opentelemetry.trace import Status, StatusCode
+
+from openlit.__helpers import (
+    general_tokens,
+    get_chat_model_cost,
+    common_span_attributes,
+    record_completion_metrics,
+)
+from openlit.semcov import SemanticConvention
+
+def get_inference_config(args, kwargs):
+    """
+    Safely extract inference configuration from args or kwargs.
+    """
+
+    if 'sampling_params' in kwargs:
+        return kwargs['sampling_params']
+    if len(args) > 1:
+        return args[1]
+    return None
+
+def format_content(prompts):
+    """
+    Process a list of prompts to extract content.
+    """
+
+    if isinstance(prompts, str):
+        return prompts
+    elif isinstance(prompts, list):
+        return "\n".join(str(prompt) for prompt in prompts)
+    else:
+        return str(prompts)
+
+def common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+    capture_message_content, disable_metrics, version, is_stream):
+    """
+    Process chat request and generate Telemetry
+    """
+
+    request_model = scope._request_model
+
+    # Extract prompts and completions from vLLM response
+    input_tokens = 0
+    output_tokens = 0
+    prompt = ""
+    completion = ""
+
+    for output in scope._response:
+        prompt += output.prompt + "\n"
+        if output.outputs and len(output.outputs) > 0:
+            completion += output.outputs[0].text + "\n"
+        input_tokens += general_tokens(output.prompt)
+        output_tokens += general_tokens(output.outputs[0].text)
+
+    cost = get_chat_model_cost(request_model, pricing_info, input_tokens, output_tokens)
+
+    # Common Span Attributes
+    common_span_attributes(scope,
+        SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT, SemanticConvention.GEN_AI_SYSTEM_VLLM,
+        scope._server_address, scope._server_port, request_model, request_model,
+        environment, application_name, is_stream, scope._tbt, scope._ttft, version)
+
+    # Span Attributes for Request parameters
+    inference_config = get_inference_config(scope._args, scope._kwargs)
+    if inference_config:
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_MAX_TOKENS, getattr(inference_config, 'max_tokens', -1))
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_STOP_SEQUENCES, getattr(inference_config, 'stop_sequences', []))
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_TEMPERATURE, getattr(inference_config, 'temperature', 1.0))
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_TOP_P, getattr(inference_config, 'top_p', 1.0))
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_TOP_K, getattr(inference_config, 'top_k', -1))
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_PRESENCE_PENALTY,
+            getattr(inference_config, 'presence_penalty', 0.0))
+        scope._span.set_attribute(SemanticConvention.GEN_AI_REQUEST_FREQUENCY_PENALTY,
+            getattr(inference_config, 'frequency_penalty', 0.0))
+
+    # Span Attributes for Response parameters
+    scope._span.set_attribute(SemanticConvention.GEN_AI_OUTPUT_TYPE, "text")
+
+    # Span Attributes for Cost and Tokens
+    scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_INPUT_TOKENS, input_tokens)
+    scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_OUTPUT_TOKENS, output_tokens)
+    scope._span.set_attribute(SemanticConvention.GEN_AI_CLIENT_TOKEN_USAGE, input_tokens + output_tokens)
+    scope._span.set_attribute(SemanticConvention.GEN_AI_USAGE_COST, cost)
+
+    # Span Attributes for Content
+    if capture_message_content:
+        scope._span.set_attribute(SemanticConvention.GEN_AI_CONTENT_PROMPT, prompt)
+        scope._span.set_attribute(SemanticConvention.GEN_AI_CONTENT_COMPLETION, completion)
+
+        # To be removed once the change to span_attributes (from span events) is complete
+        scope._span.add_event(
+            name=SemanticConvention.GEN_AI_CONTENT_PROMPT_EVENT,
+            attributes={
+                SemanticConvention.GEN_AI_CONTENT_PROMPT: prompt,
+            },
+        )
+        scope._span.add_event(
+            name=SemanticConvention.GEN_AI_CONTENT_COMPLETION_EVENT,
+            attributes={
+                SemanticConvention.GEN_AI_CONTENT_COMPLETION: completion,
+            },
+        )
+
+    scope._span.set_status(Status(StatusCode.OK))
+
+    # Metrics
+    if not disable_metrics:
+        record_completion_metrics(metrics, SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT, SemanticConvention.GEN_AI_SYSTEM_VLLM,
+            scope._server_address, scope._server_port, request_model, request_model, environment,
+            application_name, scope._start_time, scope._end_time, input_tokens, output_tokens,
+            cost, scope._tbt, scope._ttft)
+
+def process_chat_response(instance, response, request_model, pricing_info, server_port, server_address,
+    environment, application_name, metrics, start_time, span, args, kwargs,
+    capture_message_content=False, disable_metrics=False, version="1.0.0"):
+    """
+    Process chat request and generate Telemetry
+    """
+
+    # Create scope object
+    scope = type("GenericScope", (), {})()
+
+    scope._response = response
+    scope._start_time = start_time
+    scope._end_time = time.time()
+    scope._span = span
+    scope._ttft, scope._tbt = scope._end_time - scope._start_time, 0
+    scope._server_address = server_address
+    scope._server_port = server_port
+    scope._request_model = request_model
+    scope._timestamps = []
+    scope._args = args
+    scope._kwargs = kwargs
+
+    common_chat_logic(scope, pricing_info, environment, application_name, metrics,
+        capture_message_content, disable_metrics, version, is_stream=False)
+
+    return response

{openlit-1.34.12 → openlit-1.34.14}/src/openlit/instrumentation/vllm/vllm.py

@@ -2,7 +2,6 @@
 Module for monitoring vLLM API calls.
 """
 
-import logging
 import time
 from opentelemetry.trace import SpanKind
 from openlit.__helpers import (
@@ -14,11 +13,8 @@ from openlit.instrumentation.vllm.utils import (
 )
 from openlit.semcov import SemanticConvention
 
-# Initialize logger for logging potential issues and operations
-logger = logging.getLogger(__name__)
-
-def generate(version, environment, application_name,
-    tracer, pricing_info, capture_message_content, metrics, disable_metrics):
+def generate(version, environment, application_name, tracer, pricing_info,
+             capture_message_content, metrics, disable_metrics):
     """
     Generates a telemetry wrapper for GenAI function call
     """
@@ -27,7 +23,6 @@ def generate(version, environment, application_name,
         """
         Wraps the GenAI function call.
         """
-
         server_address, server_port = set_server_address_and_port(instance, "http://127.0.0.1", 443)
         request_model = instance.llm_engine.model_config.model or "facebook/opt-125m"
 
@@ -56,9 +51,9 @@ def generate(version, environment, application_name,
                     disable_metrics=disable_metrics,
                     version=version,
                 )
+
             except Exception as e:
                 handle_exception(span, e)
-                logger.error("Error in trace creation: %s", e)
 
             return response
 

openlit-1.34.12/src/openlit/instrumentation/transformers/transformers.py

@@ -1,60 +0,0 @@
-"""
-Module for monitoring HF Transformers API calls.
-"""
-
-import logging
-import time
-from opentelemetry.trace import SpanKind
-from openlit.__helpers import (
-    set_server_address_and_port
-)
-from openlit.instrumentation.transformers.utils import (
-    process_chat_response,
-)
-from openlit.semcov import SemanticConvention
-
-# Initialize logger for logging potential issues and operations
-logger = logging.getLogger(__name__)
-
-def pipeline_wrapper(version, environment, application_name,
-    tracer, pricing_info, capture_message_content, metrics, disable_metrics):
-    """
-    Generates a telemetry wrapper for GenAI function call
-    """
-
-    def wrapper(wrapped, instance, args, kwargs):
-        """
-        Wraps the GenAI function call.
-        """
-
-        server_address, server_port = set_server_address_and_port(instance, "127.0.0.1", 80)
-        request_model = instance.model.config.name_or_path
-
-        span_name = f"{SemanticConvention.GEN_AI_OPERATION_TYPE_CHAT} {request_model}"
-
-        with tracer.start_as_current_span(span_name, kind= SpanKind.CLIENT) as span:
-            start_time = time.time()
-            response = wrapped(*args, **kwargs)
-
-            response = process_chat_response(
-                    instance = instance,
-                    response=response,
-                    request_model=request_model,
-                    pricing_info=pricing_info,
-                    server_port=server_port,
-                    server_address=server_address,
-                    environment=environment,
-                    application_name=application_name,
-                    metrics=metrics,
-                    start_time=start_time,
-                    span=span,
-                    args=args,
-                    kwargs=kwargs,
-                    capture_message_content=capture_message_content,
-                    disable_metrics=disable_metrics,
-                    version=version,
-            )
-
-        return response
-
-    return wrapper