Skip to content

LLM inference in C/C++

C++ ggml

model-saver : fix expert shared/chunk FFN length key clobber (#26693)

Caleb DeLeeuw0377426cef5d
diff --git a/src/llama-model-saver.cpp b/src/llama-model-saver.cpp
index 3812c594e7..248d1ed387 100644
--- a/src/llama-model-saver.cpp
+++ b/src/llama-model-saver.cpp
@@ -213,7 +213,7 @@ void llama_model_saver::add_kv_from_model() {
add_kv(LLM_KV_FEED_FORWARD_LENGTH, hparams.n_ff_arr, true);
add_kv(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp);
add_kv(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_shexp);
- add_kv(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_chexp);
+ add_kv(LLM_KV_EXPERT_CHUNK_FEED_FORWARD_LENGTH, hparams.n_ff_chexp);
add_kv(LLM_KV_SWIGLU_CLAMP_EXP, hparams.swiglu_clamp_exp);
add_kv(LLM_KV_SWIGLU_CLAMP_SHEXP, hparams.swiglu_clamp_shexp);
add_kv(LLM_KV_USE_PARALLEL_RESIDUAL, hparams.use_par_res);
diff --git a/tests/test-llama-archs.cpp b/tests/test-llama-archs.cpp
index 0e29d221ba..6ff0d0ac1d 100644
--- a/tests/test-llama-archs.cpp
+++ b/tests/test-llama-archs.cpp
@@ -217,6 +217,7 @@ static gguf_context_ptr get_gguf_ctx(const llm_arch arch, const bool moe) {
if (moe) {
ms.add_kv(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, n_ff);
+ ms.add_kv(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, n_ff / 2); // distinct from n_ff so a saver key-clobber surfaces on reload
ms.add_kv(LLM_KV_INTERLEAVE_MOE_LAYER_STEP, uint32_t(2));
ms.add_kv(LLM_KV_EXPERT_COUNT, uint32_t(2));
ms.add_kv(LLM_KV_EXPERT_USED_COUNT, uint32_t(1));