Use n_ctx in kv find slot for consistency

2023-10-06 06:23:49 -06:00 · 2023-10-06 06:23:49 -06:00 · 3144563db1
commit 3144563db1
parent abafd01ec8
1 changed files with 1 additions and 1 deletions
--- a/llama.cpp
+++ b/llama.cpp
@ -1316,7 +1316,7 @@ static bool llama_kv_cache_find_slot(

    while (true) {
        if (cache.head + n_tokens > n_ctx) {
-            n_tested += cache.size - cache.head;
+            n_tested += n_ctx - cache.head;
            cache.head = 0;
            continue;
        }