From aa0e28f8fcc540e4c01b71ff65761551518edc27 Mon Sep 17 00:00:00 2001
From: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
Date: Sat, 20 Apr 2024 10:17:03 +0200
Subject: [PATCH] common: add llama_model_kv_override_free common: free kv
 override if used after model loading

---
 common/common.cpp | 12 ++++++++++++
 common/common.h   |  3 +++
 llama.cpp         |  8 --------
 llama.h           |  3 ---
 4 files changed, 15 insertions(+), 11 deletions(-)

diff --git a/common/common.cpp b/common/common.cpp
index 9838a538f..c30b9e59f 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -2232,6 +2232,10 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
         return std::make_tuple(nullptr, nullptr);
     }
 
+    if (!params.kv_overrides.empty()) {
+        llama_model_kv_override_free(params.kv_overrides.data());
+    }
+
     auto cparams = llama_context_params_from_gpt_params(params);
 
     llama_context * lctx = llama_new_context_with_model(model, cparams);
@@ -2952,3 +2956,11 @@ llama_control_vector_data llama_control_vector_load(const std::vector<llama_cont
 
     return result;
 }
+
+void llama_model_kv_override_free(struct llama_model_kv_override * kv_overrides) {
+    for (const struct llama_model_kv_override *p = kv_overrides; p->key[0] != 0; p++) {
+        if (p->tag == LLAMA_KV_OVERRIDE_TYPE_STR) {
+            delete p->str_value;
+        }
+    }
+}
diff --git a/common/common.h b/common/common.h
index 562d3a119..916023e25 100644
--- a/common/common.h
+++ b/common/common.h
@@ -172,6 +172,9 @@ struct gpt_params {
 
 bool parse_kv_override(const char * data, std::vector<llama_model_kv_override> & overrides);
 
+// Frees all allocated memory
+void llama_model_kv_override_free(struct llama_model_kv_override * ctx);
+
 bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params);
 
 bool gpt_params_parse(int argc, char ** argv, gpt_params & params);
diff --git a/llama.cpp b/llama.cpp
index 2907283f4..9d9f7b4e1 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -14987,14 +14987,6 @@ void llama_free_model(struct llama_model * model) {
     delete model;
 }
 
-void llama_model_kv_override_free(struct llama_model_kv_override * kv_overrides) {
-    for (const struct llama_model_kv_override *p = kv_overrides; p->key[0] != 0; p++) {
-        if (p->tag == LLAMA_KV_OVERRIDE_TYPE_STR) {
-            delete p->str_value;
-        }
-    }
-}
-
 struct llama_context * llama_new_context_with_model(
                  struct llama_model * model,
         struct llama_context_params   params) {
diff --git a/llama.h b/llama.h
index 0f215f336..0cc0d3f29 100644
--- a/llama.h
+++ b/llama.h
@@ -209,9 +209,6 @@ extern "C" {
         };
     };
 
-    // Frees all allocated memory
-    LLAMA_API void llama_model_kv_override_free(struct llama_model_kv_override * ctx);
-
     struct llama_model_params {
         int32_t n_gpu_layers; // number of layers to store in VRAM
         enum llama_split_mode split_mode; // how to split the model across multiple GPUs