Logging
Logging records request prompts, generated outputs, latency, errors, and model versions for debugging, auditing, safety review, and post-hoc analysis of LLM behaviour.
What is Logging?
Logging records request prompts, generated outputs, latency, errors, and model versions for debugging, auditing, safety review, and post-hoc analysis of LLM behaviour.
Logging records request prompts, generated outputs, latency, errors, and model versions for debugging, auditing, safety review, and post-hoc analysis of LLM behaviour.
Where is it used?
Langfuse, Helicone, and Phoenix log LLM I/O for production apps; vLLM and TGI expose structured logs; OpenAI Dashboard logs API calls for billing and abuse review.
How to build it
Integrate `langfuse` SDK by wrapping the OpenAI client with `langfuse.openai.OpenAI()`, and inspect traced prompts, responses, token counts, and latency in the Langfuse dashboard.