Skip to content

Same agent Smaller bill. The tools and infrastructure to get more work from your AI. Fewer tokens. Less compute. Lower cost.

Add Support for LangChain

patcher979434f2e7370
diff --git a/sdk/python/src/openlit/instrumentation/langchain/__init__.py b/sdk/python/src/openlit/instrumentation/langchain/__init__.py
index 805138ee..2d5037a1 100644
--- a/sdk/python/src/openlit/instrumentation/langchain/__init__.py
+++ b/sdk/python/src/openlit/instrumentation/langchain/__init__.py
@@ -5,7 +5,14 @@ import importlib.metadata
from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
from wrapt import wrap_function_wrapper
-from openlit.instrumentation.langchain.langchain import general_wrap, hub, llm, allm
+from openlit.instrumentation.langchain.langchain import (
+ general_wrap,
+ hub,
+ llm,
+ allm,
+ chat,
+ achat
+)
_instruments = ("langchain >= 0.1.20",)
@@ -52,6 +59,18 @@ WRAPPED_METHODS = [
"endpoint": "langchain.llm",
"wrapper": allm,
},
+ {
+ "package": "langchain_core.language_models.chat_models",
+ "object": "BaseChatModel.invoke",
+ "endpoint": "langchain.chat_models",
+ "wrapper": chat,
+ },
+ {
+ "package": "langchain_core.language_models.chat_models",
+ "object": "BaseChatModel.ainvoke",
+ "endpoint": "langchain.chat_models",
+ "wrapper": achat,
+ },
]
class LangChainInstrumentor(BaseInstrumentor):
@@ -66,6 +85,8 @@ class LangChainInstrumentor(BaseInstrumentor):
tracer = kwargs.get("tracer")
pricing_info = kwargs.get("pricing_info")
trace_content = kwargs.get("trace_content")
+ metrics = kwargs.get("metrics_dict")
+ disable_metrics = kwargs.get("disable_metrics")
version = importlib.metadata.version("langchain")
for wrapped_method in WRAPPED_METHODS:
@@ -77,7 +98,7 @@ class LangChainInstrumentor(BaseInstrumentor):
wrap_package,
wrap_object,
wrapper(gen_ai_endpoint, version, environment, application_name,
- tracer, pricing_info, trace_content),
+ tracer, pricing_info, trace_content, metrics, disable_metrics),
)
@staticmethod
diff --git a/sdk/python/src/openlit/instrumentation/langchain/langchain.py b/sdk/python/src/openlit/instrumentation/langchain/langchain.py
index f7b3fbce..d9613288 100644
--- a/sdk/python/src/openlit/instrumentation/langchain/langchain.py
+++ b/sdk/python/src/openlit/instrumentation/langchain/langchain.py
@@ -6,14 +6,14 @@ Module for monitoring Langchain applications.
import logging
from opentelemetry.trace import SpanKind, Status, StatusCode
from opentelemetry.sdk.resources import TELEMETRY_SDK_NAME
-from openlit.__helpers import handle_exception
+from openlit.__helpers import handle_exception, get_chat_model_cost, general_tokens
from openlit.semcov import SemanticConvetion
# Initialize logger for logging potential issues and operations
logger = logging.getLogger(__name__)
def general_wrap(gen_ai_endpoint, version, environment, application_name,
- tracer, pricing_info, trace_content):
+ tracer, pricing_info, trace_content, metrics, disable_metrics):
"""
Creates a wrapper around a function call to trace and log its execution metrics.
@@ -86,7 +86,7 @@ def general_wrap(gen_ai_endpoint, version, environment, application_name,
return wrapper
def hub(gen_ai_endpoint, version, environment, application_name, tracer,
- pricing_info, trace_content):
+ pricing_info, trace_content, metrics, disable_metrics):
"""
Creates a wrapper around Langchain hub operations for tracing and logging.
@@ -162,7 +162,7 @@ def hub(gen_ai_endpoint, version, environment, application_name, tracer,
def allm(gen_ai_endpoint, version, environment, application_name,
- tracer, pricing_info, trace_content):
+ tracer, pricing_info, trace_content, metrics, disable_metrics):
"""
Creates a wrapper around a function call to trace and log its execution metrics.
@@ -207,6 +207,16 @@ def allm(gen_ai_endpoint, version, environment, application_name,
response = await wrapped(*args, **kwargs)
try:
+ prompt = args[0] or ""
+ # input_tokens = general_tokens(prompt)
+ # output_tokens = general_tokens(response)
+
+ # # Calculate cost of the operation
+ # cost = get_chat_model_cost(
+ # str(getattr(instance, 'model')),
+ # pricing_info, input_tokens, output_tokens
+ # )
+
span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN)
@@ -226,13 +236,56 @@ def allm(gen_ai_endpoint, version, environment, application_name,
str(getattr(instance, 'top_k')))
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TOP_P,
str(getattr(instance, 'top_p')))
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_IS_STREAM,
+ False)
+ # span.set_attribute(SemanticConvetion.GEN_AI_USAGE_PROMPT_TOKENS,
+ # input_tokens)
+ # span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COMPLETION_TOKENS,
+ # output_tokens)
+ # span.set_attribute(SemanticConvetion.GEN_AI_USAGE_TOTAL_TOKENS,
+ # input_tokens + output_tokens)
+ # span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST,
+ # cost)
if trace_content:
- span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_PROMPT,
- args[0])
- span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_COMPLETION,
- response)
+ span.add_event(
+ name=SemanticConvetion.GEN_AI_CONTENT_PROMPT_EVENT,
+ attributes={
+ SemanticConvetion.GEN_AI_CONTENT_PROMPT: prompt,
+ },
+ )
+ span.add_event(
+ name=SemanticConvetion.GEN_AI_CONTENT_COMPLETION_EVENT,
+ attributes={
+ SemanticConvetion.GEN_AI_CONTENT_COMPLETION: response,
+ },
+ )
+
span.set_status(Status(StatusCode.OK))
+ # if disable_metrics is False:
+ # attributes = {
+ # TELEMETRY_SDK_NAME:
+ # "openlit",
+ # SemanticConvetion.GEN_AI_APPLICATION_NAME:
+ # application_name,
+ # SemanticConvetion.GEN_AI_SYSTEM:
+ # SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN,
+ # SemanticConvetion.GEN_AI_ENVIRONMENT:
+ # environment,
+ # SemanticConvetion.GEN_AI_TYPE:
+ # SemanticConvetion.GEN_AI_TYPE_CHAT,
+ # SemanticConvetion.GEN_AI_REQUEST_MODEL:
+ # str(getattr(instance, 'model'))
+ # }
+
+ # metrics["genai_requests"].add(1, attributes)
+ # metrics["genai_total_tokens"].add(
+ # input_tokens + output_tokens, attributes
+ # )
+ # metrics["genai_completion_tokens"].add(output_tokens, attributes)
+ # metrics["genai_prompt_tokens"].add(input_tokens, attributes)
+ # metrics["genai_cost"].record(cost, attributes)
+
# Return original response
return response
@@ -246,7 +299,7 @@ def allm(gen_ai_endpoint, version, environment, application_name,
return wrapper
def llm(gen_ai_endpoint, version, environment, application_name,
- tracer, pricing_info, trace_content):
+ tracer, pricing_info, trace_content, metrics, disable_metrics):
"""
Creates a wrapper around a function call to trace and log its execution metrics.
@@ -291,6 +344,16 @@ def llm(gen_ai_endpoint, version, environment, application_name,
response = wrapped(*args, **kwargs)
try:
+ prompt = args[0] or ""
+ # input_tokens = general_tokens(prompt)
+ # output_tokens = general_tokens(response)
+
+ # # Calculate cost of the operation
+ # cost = get_chat_model_cost(
+ # str(getattr(instance, 'model')),
+ # pricing_info, input_tokens, output_tokens
+ # )
+
span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN)
@@ -310,13 +373,56 @@ def llm(gen_ai_endpoint, version, environment, application_name,
str(getattr(instance, 'top_k')))
span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TOP_P,
str(getattr(instance, 'top_p')))
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_IS_STREAM,
+ False)
+ # span.set_attribute(SemanticConvetion.GEN_AI_USAGE_PROMPT_TOKENS,
+ # input_tokens)
+ # span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COMPLETION_TOKENS,
+ # output_tokens)
+ # span.set_attribute(SemanticConvetion.GEN_AI_USAGE_TOTAL_TOKENS,
+ # input_tokens + output_tokens)
+ # span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST,
+ # cost)
if trace_content:
- span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_PROMPT,
- args[0])
- span.set_attribute(SemanticConvetion.GEN_AI_CONTENT_COMPLETION,
- response)
+ span.add_event(
+ name=SemanticConvetion.GEN_AI_CONTENT_PROMPT_EVENT,
+ attributes={
+ SemanticConvetion.GEN_AI_CONTENT_PROMPT: prompt,
+ },
+ )
+ span.add_event(
+ name=SemanticConvetion.GEN_AI_CONTENT_COMPLETION_EVENT,
+ attributes={
+ SemanticConvetion.GEN_AI_CONTENT_COMPLETION: response,
+ },
+ )
+
span.set_status(Status(StatusCode.OK))
+ # if disable_metrics is False:
+ # attributes = {
+ # TELEMETRY_SDK_NAME:
+ # "openlit",
+ # SemanticConvetion.GEN_AI_APPLICATION_NAME:
+ # application_name,
+ # SemanticConvetion.GEN_AI_SYSTEM:
+ # SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN,
+ # SemanticConvetion.GEN_AI_ENVIRONMENT:
+ # environment,
+ # SemanticConvetion.GEN_AI_TYPE:
+ # SemanticConvetion.GEN_AI_TYPE_CHAT,
+ # SemanticConvetion.GEN_AI_REQUEST_MODEL:
+ # str(getattr(instance, 'model'))
+ # }
+
+ # metrics["genai_requests"].add(1, attributes)
+ # metrics["genai_total_tokens"].add(
+ # input_tokens + output_tokens, attributes
+ # )
+ # metrics["genai_completion_tokens"].add(output_tokens, attributes)
+ # metrics["genai_prompt_tokens"].add(input_tokens, attributes)
+ # metrics["genai_cost"].record(cost, attributes)
+
# Return original response
return response
@@ -328,3 +434,279 @@ def llm(gen_ai_endpoint, version, environment, application_name,
return response
return wrapper
+
+def chat(gen_ai_endpoint, version, environment, application_name,
+ tracer, pricing_info, trace_content, metrics, disable_metrics):
+ """
+ Creates a wrapper around a function call to trace and log its execution metrics.
+
+ This function wraps any given function to measure its execution time,
+ log its operation, and trace its execution using OpenTelemetry.
+
+ Parameters:
+ - gen_ai_endpoint (str): A descriptor or name for the endpoint being traced.
+ - version (str): The version of the Langchain application.
+ - environment (str): The deployment environment (e.g., 'production', 'development').
+ - application_name (str): Name of the Langchain application.
+ - tracer (opentelemetry.trace.Tracer): The tracer object used for OpenTelemetry tracing.
+ - pricing_info (dict): Information about the pricing for internal metrics (currently not used).
+ - trace_content (bool): Flag indicating whether to trace the content of the response.
+
+ Returns:
+ - function: A higher-order function that takes a function 'wrapped' and returns
+ a new function that wraps 'wrapped' with additional tracing and logging.
+ """
+
+ def wrapper(wrapped, instance, args, kwargs):
+ """
+ An inner wrapper function that executes the wrapped function, measures execution
+ time, and records trace data using OpenTelemetry.
+
+ Parameters:
+ - wrapped (Callable): The original function that this wrapper will execute.
+ - instance (object): The instance to which the wrapped function belongs. This
+ is used for instance methods. For static and classmethods,
+ this may be None.
+ - args (tuple): Positional arguments passed to the wrapped function.
+ - kwargs (dict): Keyword arguments passed to the wrapped function.
+
+ Returns:
+ - The result of the wrapped function call.
+
+ The wrapper initiates a span with the provided tracer, sets various attributes
+ on the span based on the function's execution and response, and ensures
+ errors are handled and logged appropriately.
+ """
+ with tracer.start_as_current_span(gen_ai_endpoint, kind= SpanKind.CLIENT) as span:
+ response = wrapped(*args, **kwargs)
+
+ try:
+ input_tokens = response.response_metadata["prompt_eval_count"] or 0
+ output_tokens = response.response_metadata["eval_count"] or 0
+
+ # Calculate cost of the operation
+ cost = get_chat_model_cost(
+ str(getattr(instance, 'model')),
+ pricing_info, input_tokens, output_tokens
+ )
+
+ span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
+ span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
+ SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN)
+ span.set_attribute(SemanticConvetion.GEN_AI_ENDPOINT,
+ gen_ai_endpoint)
+ span.set_attribute(SemanticConvetion.GEN_AI_ENVIRONMENT,
+ environment)
+ span.set_attribute(SemanticConvetion.GEN_AI_TYPE,
+ SemanticConvetion.GEN_AI_TYPE_CHAT)
+ span.set_attribute(SemanticConvetion.GEN_AI_APPLICATION_NAME,
+ application_name)
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MODEL,
+ str(getattr(instance, 'model')))
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TEMPERATURE,
+ str(getattr(instance, 'temperature')))
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TOP_K,
+ str(getattr(instance, 'top_k')))
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TOP_P,
+ str(getattr(instance, 'top_p')))
+ span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_FINISH_REASON,
+ [response.response_metadata["done_reason"]])
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_IS_STREAM,
+ False)
+ span.set_attribute(SemanticConvetion.GEN_AI_USAGE_PROMPT_TOKENS,
+ input_tokens)
+ span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COMPLETION_TOKENS,
+ output_tokens)
+ span.set_attribute(SemanticConvetion.GEN_AI_USAGE_TOTAL_TOKENS,
+ input_tokens + output_tokens)
+ span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST,
+ cost)
+ if trace_content:
+ span.add_event(
+ name=SemanticConvetion.GEN_AI_CONTENT_PROMPT_EVENT,
+ attributes={
+ SemanticConvetion.GEN_AI_CONTENT_PROMPT: args[0],
+ },
+ )
+ span.add_event(
+ name=SemanticConvetion.GEN_AI_CONTENT_COMPLETION_EVENT,
+ attributes={
+ SemanticConvetion.GEN_AI_CONTENT_COMPLETION: response.content,
+ },
+ )
+
+ span.set_status(Status(StatusCode.OK))
+
+ if disable_metrics is False:
+ attributes = {
+ TELEMETRY_SDK_NAME:
+ "openlit",
+ SemanticConvetion.GEN_AI_APPLICATION_NAME:
+ application_name,
+ SemanticConvetion.GEN_AI_SYSTEM:
+ SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN,
+ SemanticConvetion.GEN_AI_ENVIRONMENT:
+ environment,
+ SemanticConvetion.GEN_AI_TYPE:
+ SemanticConvetion.GEN_AI_TYPE_CHAT,
+ SemanticConvetion.GEN_AI_REQUEST_MODEL:
+ str(getattr(instance, 'model'))
+ }
+
+ metrics["genai_requests"].add(1, attributes)
+ metrics["genai_total_tokens"].add(
+ input_tokens + output_tokens, attributes
+ )
+ metrics["genai_completion_tokens"].add(output_tokens, attributes)
+ metrics["genai_prompt_tokens"].add(input_tokens, attributes)
+ metrics["genai_cost"].record(cost, attributes)
+
+ # Return original response
+ return response
+
+ except Exception as e:
+ handle_exception(span, e)
+ logger.error("Error in trace creation: %s", e)
+
+ # Return original response
+ return response
+
+ return wrapper
+
+def achat(gen_ai_endpoint, version, environment, application_name,
+ tracer, pricing_info, trace_content, metrics, disable_metrics):
+ """
+ Creates a wrapper around a function call to trace and log its execution metrics.
+
+ This function wraps any given function to measure its execution time,
+ log its operation, and trace its execution using OpenTelemetry.
+
+ Parameters:
+ - gen_ai_endpoint (str): A descriptor or name for the endpoint being traced.
+ - version (str): The version of the Langchain application.
+ - environment (str): The deployment environment (e.g., 'production', 'development').
+ - application_name (str): Name of the Langchain application.
+ - tracer (opentelemetry.trace.Tracer): The tracer object used for OpenTelemetry tracing.
+ - pricing_info (dict): Information about the pricing for internal metrics (currently not used).
+ - trace_content (bool): Flag indicating whether to trace the content of the response.
+
+ Returns:
+ - function: A higher-order function that takes a function 'wrapped' and returns
+ a new function that wraps 'wrapped' with additional tracing and logging.
+ """
+
+ async def wrapper(wrapped, instance, args, kwargs):
+ """
+ An inner wrapper function that executes the wrapped function, measures execution
+ time, and records trace data using OpenTelemetry.
+
+ Parameters:
+ - wrapped (Callable): The original function that this wrapper will execute.
+ - instance (object): The instance to which the wrapped function belongs. This
+ is used for instance methods. For static and classmethods,
+ this may be None.
+ - args (tuple): Positional arguments passed to the wrapped function.
+ - kwargs (dict): Keyword arguments passed to the wrapped function.
+
+ Returns:
+ - The result of the wrapped function call.
+
+ The wrapper initiates a span with the provided tracer, sets various attributes
+ on the span based on the function's execution and response, and ensures
+ errors are handled and logged appropriately.
+ """
+ with tracer.start_as_current_span(gen_ai_endpoint, kind= SpanKind.CLIENT) as span:
+ response = await wrapped(*args, **kwargs)
+
+ try:
+ input_tokens = response.response_metadata["prompt_eval_count"] or 0
+ output_tokens = response.response_metadata["eval_count"] or 0
+
+ # Calculate cost of the operation
+ cost = get_chat_model_cost(
+ str(getattr(instance, 'model')),
+ pricing_info, input_tokens, output_tokens
+ )
+
+ span.set_attribute(TELEMETRY_SDK_NAME, "openlit")
+ span.set_attribute(SemanticConvetion.GEN_AI_SYSTEM,
+ SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN)
+ span.set_attribute(SemanticConvetion.GEN_AI_ENDPOINT,
+ gen_ai_endpoint)
+ span.set_attribute(SemanticConvetion.GEN_AI_ENVIRONMENT,
+ environment)
+ span.set_attribute(SemanticConvetion.GEN_AI_TYPE,
+ SemanticConvetion.GEN_AI_TYPE_CHAT)
+ span.set_attribute(SemanticConvetion.GEN_AI_APPLICATION_NAME,
+ application_name)
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_MODEL,
+ str(getattr(instance, 'model')))
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TEMPERATURE,
+ str(getattr(instance, 'temperature')))
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TOP_K,
+ str(getattr(instance, 'top_k')))
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_TOP_P,
+ str(getattr(instance, 'top_p')))
+ span.set_attribute(SemanticConvetion.GEN_AI_RESPONSE_FINISH_REASON,
+ [response.response_metadata["done_reason"]])
+ span.set_attribute(SemanticConvetion.GEN_AI_REQUEST_IS_STREAM,
+ False)
+ span.set_attribute(SemanticConvetion.GEN_AI_USAGE_PROMPT_TOKENS,
+ input_tokens)
+ span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COMPLETION_TOKENS,
+ output_tokens)
+ span.set_attribute(SemanticConvetion.GEN_AI_USAGE_TOTAL_TOKENS,
+ input_tokens + output_tokens)
+ span.set_attribute(SemanticConvetion.GEN_AI_USAGE_COST,
+ cost)
+ if trace_content:
+ span.add_event(
+ name=SemanticConvetion.GEN_AI_CONTENT_PROMPT_EVENT,
+ attributes={
+ SemanticConvetion.GEN_AI_CONTENT_PROMPT: args[0],
+ },
+ )
+ span.add_event(
+ name=SemanticConvetion.GEN_AI_CONTENT_COMPLETION_EVENT,
+ attributes={
+ SemanticConvetion.GEN_AI_CONTENT_COMPLETION: response.content,
+ },
+ )
+
+ span.set_status(Status(StatusCode.OK))
+
+ if disable_metrics is False:
+ attributes = {
+ TELEMETRY_SDK_NAME:
+ "openlit",
+ SemanticConvetion.GEN_AI_APPLICATION_NAME:
+ application_name,
+ SemanticConvetion.GEN_AI_SYSTEM:
+ SemanticConvetion.GEN_AI_SYSTEM_LANGCHAIN,
+ SemanticConvetion.GEN_AI_ENVIRONMENT:
+ environment,
+ SemanticConvetion.GEN_AI_TYPE:
+ SemanticConvetion.GEN_AI_TYPE_CHAT,
+ SemanticConvetion.GEN_AI_REQUEST_MODEL:
+ str(getattr(instance, 'model'))
+ }
+
+ metrics["genai_requests"].add(1, attributes)
+ metrics["genai_total_tokens"].add(
+ input_tokens + output_tokens, attributes
+ )
+ metrics["genai_completion_tokens"].add(output_tokens, attributes)
+ metrics["genai_prompt_tokens"].add(input_tokens, attributes)
+ metrics["genai_cost"].record(cost, attributes)
+
+ # Return original response
+ return response
+
+ except Exception as e:
+ handle_exception(span, e)
+ logger.error("Error in trace creation: %s", e)
+
+ # Return original response
+ return response
+
+ return wrapper
\ No newline at end of file