diff --git a/src/llama-batch.cpp b/src/llama-batch.cpp
index 2700b970c9..cd9bc14fd4 100644
--- a/src/llama-batch.cpp
+++ b/src/llama-batch.cpp
@@ -714,6 +714,8 @@ llama_ubatch llama_batch_allocr::ubatch_add(const std::vector<int32_t> & idxs, u
     udata->seq_idx   .resize(LLAMA_MAX_SEQ, -1);
     udata->output    .resize(n_tokens);
 
+    udata->seq_id_data.resize(n_tokens);
+
     seq_set_t seq_set_unq;
 
     for (size_t i = 0; i < idxs.size(); ++i) {
@@ -735,7 +737,11 @@ llama_ubatch llama_batch_allocr::ubatch_add(const std::vector<int32_t> & idxs, u
         }
 
         udata->n_seq_id[i] = batch.n_seq_id[idxs[i]];
-        udata->seq_id[i]   = batch.seq_id[idxs[i]];
+        udata->seq_id_data[i].reserve(udata->n_seq_id[i]);
+        for (int s = 0; s < udata->n_seq_id[i]; ++s) {
+            udata->seq_id_data[i].push_back(batch.seq_id[idxs[i]][s]);
+        }
+        udata->seq_id[i]   = udata->seq_id_data[i].data();
         udata->output[i]   = batch.logits[idxs[i]];
 
         for (int s = 0; s < udata->n_seq_id[i]; ++s) {
diff --git a/src/llama-batch.h b/src/llama-batch.h
index db7a75b804..97213e2398 100644
--- a/src/llama-batch.h
+++ b/src/llama-batch.h
@@ -60,6 +60,8 @@ struct llama_ubatch {
         std::vector<llama_seq_id>   seq_id_unq;
         std::vector<int32_t>        seq_idx;
         std::vector<int8_t>         output;
+
+        std::vector<std::vector<llama_seq_id>> seq_id_data;
     };
 
     // the llama_ubatch pointers above point to this data if set. otherwise - points to non-owning data