From 5f722bc19559a82df160b748f26746dac7b55488 Mon Sep 17 00:00:00 2001 From: Tin Date: Sat, 19 Sep 2026 10:26:42 -0700 Subject: [PATCH] chore(router): remove in-repo escalation docs --- litellm/router_strategy/complexity_router/README.md | 13 ------------- 1 file changed, 13 deletions(-) diff --git a/litellm/router_strategy/complexity_router/README.md b/litellm/router_strategy/complexity_router/README.md index 2c2aea333f9..6505746bca1 100644 --- a/litellm/router_strategy/complexity_router/README.md +++ b/litellm/router_strategy/complexity_router/README.md @@ -68,19 +68,6 @@ still resolve to a deployment in `model_list`; this configuration does not creat - abc ``` -### Context-window escalation - -Context-window escalation is opt-in. Omit `enable_context_window_escalation` or set it to -`false` to keep the complexity-selected model without context-window replacement or filtering - -Set `enable_context_window_escalation: true` inside `complexity_router_config` to restrict the -selected tier to models whose declared windows fit the prompt, or move to the lowest configured -tier with a fitting model when none in the selected tier fit. Unknown windows do not justify -moving a request. `context_window_escalation_buffer` defaults to `0.95` - -Existing saved configurations with explicit `true` keep escalation enabled. Configurations that -omit the setting now default to disabled; set it to `true` to retain their previous behavior - ### Capability forecasting Set `classifier_type: capability` to use