docs: add release-note fragment for PR llm-d#1978 (llm-d#1985)

llm-d-router-release-notes[bot] · web-flow · commit 21f3af8fdbd1 · 2026-07-13T17:41:47.000+03:00
Signed-off-by: llm-d-router-release-notes[bot] &lt;287676111+llm-d-router-release-notes[bot]@users.noreply.github.com&gt;
Co-authored-by: llm-d-router-release-notes[bot] &lt;287676111+llm-d-router-release-notes[bot]@users.noreply.github.com&gt;
diff --git a/release-notes.d/unreleased/1978.md b/release-notes.d/unreleased/1978.md
@@ -0,0 +1,7 @@
+---
+pr: 1978
+url: https://github.com/llm-d/llm-d-router/pull/1978
+author: ezrasilvera
+date: 2026-07-13
+---
+burst-prefix-cache-producer: add a `balanceBy` parameter (`requests` (default) or `tokens`). With `tokens`, placement is balanced by prefix-block (input token) load instead of request count, charging each unit's prefix once per replica and discounting blocks it already holds, to avoid KV-saturating a single replica on high prompt-length-variance workloads. Default behavior is unchanged.
