fix: Fix off-by-one error for assertion check on layers in hybrid cache constructor

gabe-l-hart · gabe-l-hart · commit 0cc96c93141e · 2025-05-16T15:29:47.000-06:00
Branch: GraniteFour

Signed-off-by: Gabe Goodhart &lt;ghart@us.ibm.com&gt;
diff --git a/src/llama-kv-cache.cpp b/src/llama-kv-cache.cpp
@@ -2491,7 +2491,8 @@ llama_kv_cache_hybrid::llama_kv_cache_hybrid(
             }
         }
     }
-    GGML_ASSERT(max_layer == seen_layers.size());
+    LLAMA_LOG_DEBUG("max_layer=%zu, seen_layers.size()=%zu\n", max_layer, seen_layers.size());
+    GGML_ASSERT(max_layer + 1 == seen_layers.size());
 }
 
 void llama_kv_cache_hybrid::clear() {

Original file line number	Diff line number	Diff line change
`@@ -2491,7 +2491,8 @@ llama_kv_cache_hybrid::llama_kv_cache_hybrid(`
`2491`	`2491`	`}`
`2492`	`2492`	`}`
`2493`	`2493`	`}`
`2494`		`- GGML_ASSERT(max_layer == seen_layers.size());`
	`2494`	`+ LLAMA_LOG_DEBUG("max_layer=%zu, seen_layers.size()=%zu\n", max_layer, seen_layers.size());`
	`2495`	`+ GGML_ASSERT(max_layer + 1 == seen_layers.size());`
`2495`	`2496`	`}`
`2496`	`2497`
`2497`	`2498`	`void llama_kv_cache_hybrid::clear() {`