- fixed thread-safety test

- moved asserts to load_hparams

Author: sfallah

src/llama-graph.cpp

@@ -1856,6 +1856,9 @@ llm_graph_input_mem_hybrid * llm_graph_context::build_inp_mem_hybrid() const {
 void llm_graph_context::build_dense_out(
     ggml_tensor * dense_2,
     ggml_tensor * dense_3) const {
+    if (!cparams.embeddings) {
+        return;
+    }
     ggml_tensor * cur = res->get_embd_pooled();
     if (dense_2 != nullptr) {
         cur = ggml_mul_mat(ctx0, dense_2, cur);

src/llama-model.cpp

@@ -1230,6 +1230,9 @@ void llama_model::load_hparams(llama_model_loader & ml) {
                 ml.get_key(LLM_KV_DENSE_3_FEAT_IN, hparams.dense_3_feat_in, false);
                 ml.get_key(LLM_KV_DENSE_3_FEAT_OUT, hparams.dense_3_feat_out, false);
 
+                GGML_ASSERT((hparams.dense_2_feat_in == 0 || hparams.dense_2_feat_in == hparams.n_embd) && "dense_2_feat_in must be equal to n_embd");
+                GGML_ASSERT((hparams.dense_3_feat_out == 0 || hparams.dense_3_feat_out == hparams.n_embd) && "dense_3_feat_out must be equal to n_embd");
+
                 switch (hparams.n_layer) {
                     case 24: type = LLM_TYPE_0_3B; break;
                     default: type = LLM_TYPE_UNKNOWN;
@@ -3678,9 +3681,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
 
                     // Dense linear weights
                     dense_2_out_layers = create_tensor(tn(LLM_TENSOR_DENSE_2_OUT, "weight"), {n_embd, hparams.dense_2_feat_out}, TENSOR_NOT_REQUIRED);
-                    GGML_ASSERT((dense_2_out_layers == nullptr || hparams.dense_2_feat_in == n_embd) && "dense_2_feat_in must be equal to n_embd");
                     dense_3_out_layers = create_tensor(tn(LLM_TENSOR_DENSE_3_OUT, "weight"), {hparams.dense_3_feat_in, n_embd}, TENSOR_NOT_REQUIRED);
-                    GGML_ASSERT((dense_3_out_layers == nullptr || hparams.dense_3_feat_out == n_embd) && "dense_3_feat_out must be equal to n_embd");
 
 
                     for (int i = 0; i < n_layer; ++i) {