LangChain¶
Use DDTrace to obtain cost estimates, prompt and completion sampling, error tracking, performance Metrics, etc., from requests in the LangChain Python library.
Configuration¶
Install DDTrace¶
pip install ddtrace>=1.17
DataKit Configuration¶
- Enable the DDTrace collector
The DDTrace collector is used for collecting tracing information. Go to the DataKit installation directory and execute conf.d/ddtrace/, copy ddtrace.conf.sample and rename it to ddtrace.conf.
- Enable the StatsD collector
The StatsD collector is used for collecting Metrics information. The default port is 8125.
- Restart DataKit
Run the application¶
DD_SERVICE="my-langchain" DD_ENV="dev" DD_AGENT_HOST="localhost" DD_AGENT_PORT="9529" ddtrace-run python <your-app>.py
If you need to enable debug mode, add the parameter --debug when starting:
DD_SERVICE="my-langchain" DD_ENV="dev" DD_AGENT_HOST="localhost" DD_AGENT_PORT="9529" ddtrace-run --debug python <your-app>.py
Metrics¶
langchain¶
| Metrics | Units | Description |
|---|---|---|
| request_duration | nanoseconds | Distribution of request duration. |
| request_error | errors | Number of request exceptions. |
| tokens_completion | tokens/request | Number of tokens used in the response upon completion. |
| tokens_prompt | tokens/request | Number of tokens used in the request prompt. |
| tokens_total | tokens/request | Total number of tokens used in both request and response. |
| tokens_total_cost | dollars | Estimated cost (in dollars) based on token usage. |