docs: estimate

Signed-off-by: thxCode <[email protected]>
gpustack · Jun 14, 2024 · 85cfcd0 · 85cfcd0
1 parent f40a734
commit 85cfcd0
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/file_estimate.go b/file_estimate.go
@@ -315,8 +315,10 @@ func (gf *GGUFFile) EstimateLLaMACppUsage(opts ...LLaMACppUsageEstimateOption) (
 					rs := l.Bytes()
 					attnInc += rs
 				}
+				// https://github.com/ggerganov/llama.cpp/blob/172c8256840ffd882ab9992ecedbb587d9b21f15/llama.cpp#L6986-L6992.
 				rs := o.CacheKeyType.RowSizeOf([]uint64{uint64(a.AttentionKeyLength), nKV, a.AttentionHeadCountKV})
 				attnInc += rs
+				// https://github.com/ggerganov/llama.cpp/blob/172c8256840ffd882ab9992ecedbb587d9b21f15/llama.cpp#L7000-L7007.
 				rs = o.CacheValueType.RowSizeOf([]uint64{uint64(a.AttentionValueLength), nKV, a.AttentionHeadCountKV})
 				attnInc += rs
 			} else {