diff --git a/crates/llama_cpp/src/model/mod.rs b/crates/llama_cpp/src/model/mod.rs
index c5e4b23..29487d3 100644
--- a/crates/llama_cpp/src/model/mod.rs
+++ b/crates/llama_cpp/src/model/mod.rs
@@ -384,6 +384,7 @@ impl LlamaModel {
                 token.0,
                 buffer.as_mut_ptr() as *mut c_char,
                 std::os::raw::c_int::from(initial_size),
+                true,
             )
         };
 
@@ -398,6 +399,7 @@ impl LlamaModel {
                     token.0,
                     buffer.as_mut_ptr() as *mut c_char,
                     std::os::raw::c_int::from(buffer.len() as i32),
+                    true,
                 )
             };
             assert_eq!(size as usize, buffer.len(), "Buffer length doesn't match");
@@ -442,6 +444,7 @@ impl LlamaModel {
                     t.0,
                     token_buf.as_mut_ptr() as *mut c_char,
                     token_buf.len() as i32,
+                    true,
                 )
             };
 
diff --git a/crates/llama_cpp/src/model/params.rs b/crates/llama_cpp/src/model/params.rs
index ed4abc6..5d20995 100644
--- a/crates/llama_cpp/src/model/params.rs
+++ b/crates/llama_cpp/src/model/params.rs
@@ -112,6 +112,8 @@ impl From<LlamaParams> for llama_model_params {
             vocab_only: value.vocab_only,
             use_mmap: value.use_mmap,
             use_mlock: value.use_mlock,
+            check_tensors: false,
+            rpc_servers: ptr::null_mut(),
         }
     }
 }
diff --git a/crates/llama_cpp/src/session/params.rs b/crates/llama_cpp/src/session/params.rs
index d678968..e234bc3 100644
--- a/crates/llama_cpp/src/session/params.rs
+++ b/crates/llama_cpp/src/session/params.rs
@@ -349,6 +349,7 @@ impl From<SessionParams> for llama_context_params {
             pooling_type: value.pooling.into(),
             abort_callback: None,
             abort_callback_data: null_mut(),
+            flash_attn: false,
         }
     }
 }
diff --git a/crates/llama_cpp_sys/build.rs b/crates/llama_cpp_sys/build.rs
index a00a07e..538498c 100644
--- a/crates/llama_cpp_sys/build.rs
+++ b/crates/llama_cpp_sys/build.rs
@@ -495,7 +495,7 @@ fn compile_metal(cx: &mut Build, cxx: &mut Build) {
     let common = LLAMA_PATH.join("ggml-common.h");
 
     let input_file = File::open(ggml_metal_shader_path).expect("Failed to open input file");
-    let mut output_file = File::create(&ggml_metal_shader_out_path).expect("Failed to create output file");
+    let output_file = File::create(&ggml_metal_shader_out_path).expect("Failed to create output file");
 
     let output = Command::new("sed")
         .arg("-e")
diff --git a/crates/llama_cpp_sys/include/build-info.h b/crates/llama_cpp_sys/include/build-info.h
index 17561f1..4defbfd 100644
--- a/crates/llama_cpp_sys/include/build-info.h
+++ b/crates/llama_cpp_sys/include/build-info.h
@@ -13,7 +13,7 @@
 #ifndef BUILD_INFO_H
 #define BUILD_INFO_H
 
-#define BUILD_NUMBER 2589
-#define BUILD_COMMIT "60cdf40"
+#define BUILD_NUMBER 3038
+#define BUILD_COMMIT "fb76ec3"
 
 #endif // BUILD_INFO_H
diff --git a/crates/llama_cpp_sys/thirdparty/llama.cpp b/crates/llama_cpp_sys/thirdparty/llama.cpp
index 60cdf40..fb76ec3 160000
--- a/crates/llama_cpp_sys/thirdparty/llama.cpp
+++ b/crates/llama_cpp_sys/thirdparty/llama.cpp
@@ -1 +1 @@
-Subproject commit 60cdf40cc32f0ad4cb11e0ca8fd38f3b93d8d640
+Subproject commit fb76ec31a9914b7761c1727303ab30380fd4f05c