[Doc] Fix group numbering in Case 3 of hybrid_kv_cache_manager.md (#52160)
Signed-off-by: qwerqwerqwe8688-jpg <xuuuuu2021@163.com>
QWERQWERQWE86 committed
Aug 19, 2026 at 19:38 UTC
58302b459198eec60fad49af3c506ec10d6af821
1 file changed
+4
-3
docs/design/hybrid_kv_cache_manager.md
+4
-3
@@ -115,9 +115,10 @@ Unfortunately, not all models have such a beautiful ratio, and approach in Case
115
- Group 0: 10 full attention layers (full.0 - full.9)
116
- Group 1: 10 sliding window attention layers (sw.0 - sw.9)
117
- Group 2: 10 sliding window attention layers (sw.10 - sw.19)
118
-- ...
119
-- Group 6: 10 sliding window attention layers (sw.40 - sw.49)
120
-- Group 7: 2 sliding window attention layers (sw.50 - sw.51) and 8 padding layers
118
+- Group 3: 10 sliding window attention layers (sw.20 - sw.29)
119
+- Group 4: 10 sliding window attention layers (sw.30 - sw.39)
120
+- Group 5: 10 sliding window attention layers (sw.40 - sw.49)
121
+- Group 6: 2 sliding window attention layers (sw.50 - sw.51) and 8 padding layers
122
123
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).
124