context : fix pos_min initialization upon error decode (#14008)

ggml-ci

diff --git a/src/llama-context.cpp b/src/llama-context.cpp
index 7c1a642c19464f62b940bf9492a37dec0271e262..f1b43b9ccaaafa95abf68173972e42db59b3012f 100644 (file)
--- a/src/llama-context.cpp
+++ b/src/llama-context.cpp
@@ -1058,7 +1058,10 @@ int llama_context::decode(llama_batch & inp_batch) {
 
         if (!res) {
             // the last ubatch failed or was aborted -> remove all positions of that ubatch from the KV cache
-            llama_pos pos_min[LLAMA_MAX_PARALLEL_SEQUENCES] = { std::numeric_limits<llama_pos>::max() };
+            llama_pos pos_min[LLAMA_MAX_PARALLEL_SEQUENCES];
+            for (int s = 0; s < LLAMA_MAX_PARALLEL_SEQUENCES; ++s) {
+                pos_min[s] = std::numeric_limits<llama_pos>::max();
+            }
 
             for (uint32_t i = 0; i < ubatch.n_tokens; ++i) {
                 const auto & seq_id = ubatch.seq_id[i][0];