Skip to content

LLM inference in C/C++

C++ ggml

spec : avoid binding reference to null pointer (#27404)

Georgi Gerganovf466cfa38fac
diff --git a/common/speculative.cpp b/common/speculative.cpp
index ae55e357d5..89e9b2782c 100644
--- a/common/speculative.cpp
+++ b/common/speculative.cpp
@@ -2649,6 +2649,10 @@ void common_speculative_draft(common_speculative * spec) {
for (llama_seq_id seq_id = 0; seq_id < (llama_seq_id) dparams.size(); ++seq_id) {
auto & dp = dparams[seq_id];
+ if (!dp.drafting) {
+ continue;
+ }
+
auto & result = *dp.result;
// a new draft has been sampled