GenAI: how to record embeddings

[lmolkova]

• Which request parameters are important (in addition to model)?

  • encoding_format?
  • dimensions?
  • should they be in gen_ai.request namespace and mixed with chat params?

• Should we record input in an event?

  • embeddings calls are probably much cheaper and frequent than chats, recording input for each one could be too expensive
  • if we record inputs, they should be enabled independently from chat events