Same agent Smaller bill. The tools and infrastructure to get more work from your AI. Fewer tokens. Less compute. Lower cost.
TypeScript agent-harnessagent-harness-engineeringai-engineeringai-observabilityamd-gpuclaude-codeclickhousecodexcoding-agentcoding-agentscursorgenaigpu-monitoringllm-observabilityllmopsnvidia-smiobservabilityopen-sourceopenaiopentelemetry
[Fix]: Token usage when using chains in Langchain (#538)
Bassel-Elwakil632a66a9dc46
diff --git a/sdk/python/pyproject.toml b/sdk/python/pyproject.tomlindex 8add3047..caca1ab0 100644--- a/sdk/python/pyproject.toml+++ b/sdk/python/pyproject.toml@@ -1,6 +1,6 @@[tool.poetry]name = "openlit"-version = "1.32.7"+version = "1.32.8"description = "OpenTelemetry-native Auto instrumentation library for monitoring LLM Applications and GPUs, facilitating the integration of observability into your GenAI-driven projects"authors = ["OpenLIT"]repository = "https://github.com/openlit/openlit/tree/main/openlit/python"diff --git a/sdk/python/src/openlit/instrumentation/langchain/__init__.py b/sdk/python/src/openlit/instrumentation/langchain/__init__.pyindex 00449f3a..d501275f 100644--- a/sdk/python/src/openlit/instrumentation/langchain/__init__.py+++ b/sdk/python/src/openlit/instrumentation/langchain/__init__.py@@ -71,6 +71,12 @@ WRAPPED_METHODS = ["endpoint": "langchain.chat_models","wrapper": achat,},+ {+ "package": "langchain.chains.base",+ "object": "Chain.invoke",+ "endpoint": "langchain.chain.invoke",+ "wrapper": chat,+ }]class LangChainInstrumentor(BaseInstrumentor):diff --git a/sdk/python/src/openlit/instrumentation/langchain/langchain.py b/sdk/python/src/openlit/instrumentation/langchain/langchain.pyindex 9958fc24..32279e83 100644--- a/sdk/python/src/openlit/instrumentation/langchain/langchain.py+++ b/sdk/python/src/openlit/instrumentation/langchain/langchain.py@@ -1,4 +1,4 @@-# pylint: disable=duplicate-code, broad-exception-caught, too-many-statements, unused-argument, unused-import+# pylint: disable=duplicate-code, broad-exception-caught, too-many-statements, unused-argument, unused-import, too-many-function-args"""Module for monitoring Langchain applications."""@@ -507,12 +507,25 @@ def chat(gen_ai_endpoint, version, environment, application_name,response = wrapped(*args, **kwargs)try:- token_usage = response.response_metadata.get("token_usage", {})- input_tokens = token_usage.get("prompt_tokens", 0)- output_tokens = token_usage.get("completion_tokens", 0)- model = response.response_metadata.get("model_name", "gpt-4")-- prompt = "" if isinstance(args[0], list) else args[0]+ prompt = ""+ if hasattr(response, 'response_metadata') and response.response_metadata:+ token_usage = response.response_metadata.get("token_usage", {})+ input_tokens = token_usage.get("prompt_tokens", 0)+ output_tokens = token_usage.get("completion_tokens", 0)+ model = response.response_metadata.get("model_name", "gpt-4")+ prompt = "" if isinstance(args[0], list) else args[0]+ else:+ if not isinstance(response, dict) or "output_text" not in response:+ return response+ # Fallback: Calculate tokens manually if response_metadata is missing+ model = "gpt-4o-mini" # Fallback model+ input_texts = [+ doc.page_content for doc in response.get("input_documents", [])+ if isinstance(doc.page_content, str)+ ]+ input_tokens = sum(general_tokens(text, model) for text in input_texts)+ output_text = response.get("output_text", "")+ output_tokens = general_tokens(output_text, model)# Calculate cost of the operationcost = get_chat_model_cost(@@ -556,10 +569,11 @@ def chat(gen_ai_endpoint, version, environment, application_name,SemanticConvetion.GEN_AI_CONTENT_PROMPT: prompt,},)+ completion_content = getattr(response, 'content', "")span.add_event(name=SemanticConvetion.GEN_AI_CONTENT_COMPLETION_EVENT,attributes={- SemanticConvetion.GEN_AI_CONTENT_COMPLETION: response.content,+ SemanticConvetion.GEN_AI_CONTENT_COMPLETION: completion_content,},)diff --git a/sdk/python/tests/test_multion.py b/sdk/python/tests/test_multion.py.holdsimilarity index 100%rename from sdk/python/tests/test_multion.pyrename to sdk/python/tests/test_multion.py.hold