Skip to content

LLM inference in C/C++

C++ ggml

common : fix draft-mtp with embeddings (#26352, #27299) (#27400)

Shahir BIn Zulfiker2c6b141efb3b
diff --git a/common/speculative.cpp b/common/speculative.cpp
index 89e9b2782c..8461e4107d 100644
--- a/common/speculative.cpp
+++ b/common/speculative.cpp
@@ -2322,6 +2322,9 @@ common_params common_base_params_to_speculative(const common_params & params) {
const auto & params_spec = params.speculative.draft;
common_params result = params;
+ result.embedding = false;
+ result.pooling_type = LLAMA_POOLING_TYPE_UNSPECIFIED;
+
if (has_draft) {
result.devices = params_spec.devices;
result.model = params_spec.mparams;