diff --git a/crates/llama-cpp-bindings/src/engine.cc b/crates/llama-cpp-bindings/src/engine.cc
index a314a6a..5b48090 100644
--- a/crates/llama-cpp-bindings/src/engine.cc
+++ b/crates/llama-cpp-bindings/src/engine.cc
@@ -246,9 +246,14 @@ class TextInferenceEngineImpl : public TextInferenceEngine {
         }
 
         if (request.multibyte_pending == 0) {
-          rust::String generated_text = is_eos ? "" : request.generated_text;
-          result.push_back({request.id, generated_text});
+          rust::String generated_text;
+          try {
+            generated_text = is_eos ? "" : request.generated_text;
+          } catch (const std::invalid_argument& e) {
+            fprintf(stderr, "%s:%d [%s] - ignoring non utf-8/utf-16 output\n", __FILE__, __LINE__, __func__);
+          }
 
+          result.push_back({request.id, generated_text});
           request.generated_text.clear();
         }
       }
diff --git a/crates/llama-cpp-bindings/src/lib.rs b/crates/llama-cpp-bindings/src/lib.rs
index 5bd33d1..b2b2a32 100644
--- a/crates/llama-cpp-bindings/src/lib.rs
+++ b/crates/llama-cpp-bindings/src/lib.rs
@@ -76,7 +76,9 @@ impl AsyncTextInferenceEngine {
 
         let result = match engine.as_mut().unwrap().step() {
             Ok(result) => result,
-            Err(err) => panic!("Failed to step: {}", err),
+            Err(err) => {
+                panic!("Failed to step: {}", err)
+            }
         };
 
         for ffi::StepOutput { request_id, text } in result {