Grokking LLM Evals
Vote

0% completed

​

Logging Traces for Evaluation

The trace log of the job ticket assistant makes four promises:

  • Every request can be run again as it first ran, because the log records the input, the prompt version, the model and its settings.
  • Every output is stored as the assistant returned it, before a dispatcher edits anything.
  • Every later correction, by a dispatcher or a customer, is linked to the request it corrects.
  • Every grade, from a judge or a person, is attached to the trace it grades.

It also makes two non-promises

.....

.....

.....

Like the course? Get enrolled and start learning!

Reading Progress

0%


Vote for new content