diff --git a/Cargo.lock b/Cargo.lock
index 1334ea8..1e2c410 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2883,7 +2883,7 @@ checksum = "01cda141df6706de531b6c46c3a33ecca755538219bd484262fa09410c13539c"
 [[package]]
 name = "llama_cpp"
 version = "0.3.1"
-source = "git+https://github.com/edgenai/llama_cpp-rs?branch=main#9c098a769785738deb5e0179ad2c5e5554d1cb43"
+source = "git+https://github.com/edgenai/llama_cpp-rs?branch=main#491f69a868a3fb6169501359870bb0649f45069f"
 dependencies = [
  "derive_more",
  "futures",
@@ -2897,7 +2897,7 @@ dependencies = [
 [[package]]
 name = "llama_cpp_sys"
 version = "0.3.1"
-source = "git+https://github.com/edgenai/llama_cpp-rs?branch=main#9c098a769785738deb5e0179ad2c5e5554d1cb43"
+source = "git+https://github.com/edgenai/llama_cpp-rs?branch=main#491f69a868a3fb6169501359870bb0649f45069f"
 dependencies = [
  "ash",
  "bindgen",
diff --git a/crates/edgen_rt_llama_cpp/src/lib.rs b/crates/edgen_rt_llama_cpp/src/lib.rs
index a4ff332..1ca1874 100644
--- a/crates/edgen_rt_llama_cpp/src/lib.rs
+++ b/crates/edgen_rt_llama_cpp/src/lib.rs
@@ -259,7 +259,9 @@ impl UnloadingModel {
                 .map_err(move |e| LLMEndpointError::Advance(e.to_string()))?;
 
             let sampler = StandardSampler::default();
-            let handle = session.start_completing_with(sampler, SINGLE_MESSAGE_LIMIT);
+            let handle = session
+                .start_completing_with(sampler, SINGLE_MESSAGE_LIMIT)
+                .map_err(|e| LLMEndpointError::Advance(e.to_string()))?;
 
             Ok(handle.into_string_async().await)
         } else {
@@ -276,7 +278,9 @@ impl UnloadingModel {
                 id.advance(new_context);
 
                 let sampler = StandardSampler::default();
-                let handle = session_guard.start_completing_with(sampler, SINGLE_MESSAGE_LIMIT);
+                let handle = session_guard
+                    .start_completing_with(sampler, SINGLE_MESSAGE_LIMIT)
+                    .map_err(|e| LLMEndpointError::Advance(e.to_string()))?;
 
                 (session_signal, handle)
             };
@@ -581,7 +585,9 @@ impl CompletionStream {
 
             (
                 session_signal,
-                session_guard.start_completing_with(sampler, SINGLE_MESSAGE_LIMIT),
+                session_guard
+                    .start_completing_with(sampler, SINGLE_MESSAGE_LIMIT)
+                    .map_err(|e| LLMEndpointError::Advance(e.to_string()))?,
             )
         };
 
@@ -605,7 +611,9 @@ impl CompletionStream {
             .advance_context_async(new_context)
             .await
             .map_err(move |e| LLMEndpointError::Advance(e.to_string()))?;
-        let handle = session.start_completing_with(sampler, SINGLE_MESSAGE_LIMIT);
+        let handle = session
+            .start_completing_with(sampler, SINGLE_MESSAGE_LIMIT)
+            .map_err(|e| LLMEndpointError::Advance(e.to_string()))?;
 
         Ok(Self {
             handle: handle.into_strings(),