diff --git a/src/llama-context.cpp b/src/llama-context.cpp
index 220240ea952b..edbefcaecda3 100644
--- a/src/llama-context.cpp
+++ b/src/llama-context.cpp
@@ -194,6 +194,14 @@ llama_context::llama_context(
         cparams.causal_attn = params.attention_type == LLAMA_ATTENTION_TYPE_CAUSAL;
     }
 
+    // Non-causal models (e.g. embedding/classification encoders such as BERT) cannot generate text.
+    // A generation request (bounded n_outputs_max) would otherwise trip GGML_ASSERT in output_reserve();
+    // fail early with a clear message instead of aborting.
+    if (!cparams.causal_attn && !llama_model_has_encoder(&model) && params.n_outputs_max != 0) {
+        throw std::runtime_error("this model is non-causal (e.g. an embedding/classification model) "
+            "and cannot be used for text generation; use the embedding API (e.g. llama-embedding) instead");
+    }
+
     cparams.flash_attn = params.flash_attn_type != LLAMA_FLASH_ATTN_TYPE_DISABLED;
     cparams.auto_fa    = params.flash_attn_type == LLAMA_FLASH_ATTN_TYPE_AUTO;
 
diff --git a/tools/server/server-context.cpp b/tools/server/server-context.cpp
index 39b7eb218e69..1b3ca45305d2 100644
--- a/tools/server/server-context.cpp
+++ b/tools/server/server-context.cpp
@@ -1166,6 +1166,11 @@ struct server_context_impl {
             return false;
         }
 
+        if (ctx_tgt == nullptr) {
+            SRV_ERR("failed to create context with model, '%s'\n", params_base.model.path.c_str());
+            return false;
+        }
+
         vocab = llama_model_get_vocab(model_tgt);
 
         n_ctx = llama_n_ctx(ctx_tgt);