From 58302b459198eec60fad49af3c506ec10d6af821 Mon Sep 17 00:00:00 2001 From: QWERQWERQWE86 Date: Wed, 19 Aug 2026 19:38:34 +0800 Subject: [PATCH] [Doc] Fix group numbering in Case 3 of hybrid_kv_cache_manager.md (#52160) Signed-off-by: qwerqwerqwe8688-jpg --- docs/design/hybrid_kv_cache_manager.md | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/docs/design/hybrid_kv_cache_manager.md b/docs/design/hybrid_kv_cache_manager.md index 82d54e9b5c..9891c82f97 100644 --- a/docs/design/hybrid_kv_cache_manager.md +++ b/docs/design/hybrid_kv_cache_manager.md @@ -115,9 +115,10 @@ Unfortunately, not all models have such a beautiful ratio, and approach in Case - Group 0: 10 full attention layers (full.0 - full.9) - Group 1: 10 sliding window attention layers (sw.0 - sw.9) - Group 2: 10 sliding window attention layers (sw.10 - sw.19) -- ... -- Group 6: 10 sliding window attention layers (sw.40 - sw.49) -- Group 7: 2 sliding window attention layers (sw.50 - sw.51) and 8 padding layers +- Group 3: 10 sliding window attention layers (sw.20 - sw.29) +- Group 4: 10 sliding window attention layers (sw.30 - sw.39) +- Group 5: 10 sliding window attention layers (sw.40 - sw.49) +- Group 6: 2 sliding window attention layers (sw.50 - sw.51) and 8 padding layers We will update this algorithm if this heuristic leads to a bad result when a new model comes out (e.g., 20 full + 30 sw, the group size should be 10 instead of 20).