kv-cache : add comment + warning [no ci]

Author: ggerganov

src/llama-kv-cache.cpp

@@ -481,6 +481,12 @@ void llama_kv_cache_unified::restore() {
 }
 
 void llama_kv_cache_unified::commit() {
+    if (pending.ranges.empty()) {
+        LLAMA_LOG_WARN("%s: no pending KV cache updates to commit - might indicate a bug (ref: %s)\n",
+                __func__, "https://github.com/ggml-org/llama.cpp/pull/12695");
+        return;
+    }
+
     pending.ranges.clear();
 }
 

src/llama-kv-cache.h

@@ -144,6 +144,7 @@ class llama_kv_cache_unified : public llama_kv_cache {
         uint32_t c1 = 0;
     };
 
+    // pending cell updates that are not yet committed
     struct {
         std::vector<slot_range> ranges;
     } pending;