gracefully handle bad grammar instead of crashing

expose.cpp

@@ -195,7 +195,18 @@ extern "C"
 
     generation_outputs generate(const generation_inputs inputs)
     {
-        return gpttype_generate(inputs);
+        try {
+            return gpttype_generate(inputs);
+        } catch (const std::exception & e) {
+            generation_outputs output;
+            printf("\nGeneration encountered an exception: %s\n", e.what());
+            output.text = nullptr;
+            output.status = 0;
+            output.prompt_tokens = output.completion_tokens = 0;
+            output.stopreason = stop_reason::ERROR_ENCOUNTERED;
+            generation_finished = true;
+            return output;
+        }
     }
 
     bool sd_load_model(const sd_load_model_inputs inputs)

gpttype_adapter.cpp

@@ -789,6 +789,10 @@ static speculative_draft_result speculative_decoding_eval_chunk(llama_context *
 
 // KCPP SAMPLING FUNCTIONS
 void sample_softmax(llama_token_data_array * cur_p, bool do_sort=true) {
+    if(!(cur_p->size > 0))
+    {
+        throw std::runtime_error("No valid candidates during sampling. Current request aborted!");
+    }
     GGML_ASSERT(cur_p->size > 0);
     // Sort the logits in descending order
     if (!cur_p->sorted && do_sort) {

src/llama-model-loader.cpp

@@ -1282,7 +1282,7 @@ void llama_model_loader::done_getting_tensors() const {
         throw std::runtime_error(format("%s: wrong number of tensors; expected %d, got %d", __func__, n_tensors, n_created));
     }
     if (n_tensors_moved > 0) {
-        LLAMA_LOG_DEBUG("%s: tensor '%s' (%s) (and %zu others) cannot be used with preferred buffer type %s, using %s instead\n",
+        LLAMA_LOG_DEBUG("%s: tensor '%s' (%s) (and %zu others) moved from %s, using %s instead\n",
             __func__, first_tensor_moved_name.c_str(), first_tensor_moved_type_name.c_str(), n_tensors_moved - 1,
             ggml_backend_buft_name(first_moved_from_buft), ggml_backend_buft_name(first_moved_to_buft));
     }