From 7330d75de6b7faacd1443d0b1e46f43d00abf8f3 Mon Sep 17 00:00:00 2001
From: Meng Zhang <meng@tabbyml.com>
Date: Sun, 29 Oct 2023 16:30:16 -0700
Subject: [PATCH] chore: clear cache when there's no active requests

---
 crates/llama-cpp-bindings/llama.cpp     | 2 +-
 crates/llama-cpp-bindings/src/engine.cc | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)

diff --git a/crates/llama-cpp-bindings/llama.cpp b/crates/llama-cpp-bindings/llama.cpp
index 638ff1a..f858db8 160000
--- a/crates/llama-cpp-bindings/llama.cpp
+++ b/crates/llama-cpp-bindings/llama.cpp
@@ -1 +1 @@
-Subproject commit 638ff1aba1fa200f0bdc0ee3709176ddd783a49d
+Subproject commit f858db8db3a98968ad3764c409e43e44c443079b
diff --git a/crates/llama-cpp-bindings/src/engine.cc b/crates/llama-cpp-bindings/src/engine.cc
index 9a93f36..375553a 100644
--- a/crates/llama-cpp-bindings/src/engine.cc
+++ b/crates/llama-cpp-bindings/src/engine.cc
@@ -100,6 +100,7 @@ class TextInferenceEngineImpl : public TextInferenceEngine {
     stopped_requests_.clear();
 
     if (requests_.size() == 0) {
+      llama_kv_cache_clear(ctx);
       return {};
     }