diff --git a/.changes/unreleased/Fixed and Improvements-20241126-151903.yaml b/.changes/unreleased/Fixed and Improvements-20241126-151903.yaml
new file mode 100644
index 000000000000..2c1b320ac64b
--- /dev/null
+++ b/.changes/unreleased/Fixed and Improvements-20241126-151903.yaml	
@@ -0,0 +1,3 @@
+kind: Fixed and Improvements
+body: Fixed a panic that occurred when specifying a local model ([#3464](https://github.com/TabbyML/tabby/issues/3464))
+time: 2024-11-26T15:19:03.863207+08:00
diff --git a/crates/llama-cpp-server/src/lib.rs b/crates/llama-cpp-server/src/lib.rs
index 9b9bcb18db55..ef1ba84804d4 100644
--- a/crates/llama-cpp-server/src/lib.rs
+++ b/crates/llama-cpp-server/src/lib.rs
@@ -10,7 +10,7 @@ use serde::Deserialize;
 use supervisor::LlamaCppSupervisor;
 use tabby_common::{
     config::{HttpModelConfigBuilder, LocalModelConfig, ModelConfig},
-    registry::{parse_model_id, ModelRegistry},
+    registry::{parse_model_id, ModelRegistry, GGML_MODEL_PARTITIONED_PREFIX},
 };
 use tabby_inference::{ChatCompletionStream, CompletionOptions, CompletionStream, Embedding};
 
@@ -277,10 +277,20 @@ pub async fn create_embedding(config: &ModelConfig) -> Arc<dyn Embedding> {
 }
 
 async fn resolve_model_path(model_id: &str) -> String {
-    let (registry, name) = parse_model_id(model_id);
-    let registry = ModelRegistry::new(registry).await;
-    let path = registry.get_model_entry_path(name);
-    path.unwrap().display().to_string()
+    let path = PathBuf::from(model_id);
+    let path = if path.exists() {
+        path.join("ggml").join(format!(
+            "{}00001.gguf",
+            GGML_MODEL_PARTITIONED_PREFIX.to_owned()
+        ))
+    } else {
+        let (registry, name) = parse_model_id(model_id);
+        let registry = ModelRegistry::new(registry).await;
+        registry
+            .get_model_entry_path(name)
+            .expect("Model not found")
+    };
+    path.display().to_string()
 }
 
 #[derive(Deserialize)]