Exploring Strategies for Summary Evaluations and Context Management

Hey folks! I’ve been working a lot on Summary Evals with a number of teams, just did the session on Evals. Overflowing context windows is probably the biggest challenge in deciding how to summarize, wondering what most teams do? Map reduce chains, refine chains or no chains, etc. Let me know your thoughts, we are curious to know what the folks who are also on the ground are encountering!