From 4bd26644bf690ac482d9c60bced2228e8ad7948a Mon Sep 17 00:00:00 2001
From: Pierrick HYMBERT <pierrick.hymbert@gmail.com>
Date: Sat, 20 Apr 2024 13:24:58 +0200
Subject: [PATCH] llama: finally move the string KV override value to the stack

---
 common/common.cpp              | 14 +-------------
 common/common.h                |  3 ---
 examples/quantize/quantize.cpp |  4 ++--
 llama.h                        |  2 +-
 4 files changed, 4 insertions(+), 19 deletions(-)

diff --git a/common/common.cpp b/common/common.cpp
index c30b9e59f..f07b4d1a4 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -266,7 +266,7 @@ bool parse_kv_override(const char * data, std::vector<llama_model_kv_override> &
     } else if (strncmp(sep, "str:", 4) == 0) {
         sep += 4;
         kvo.tag = LLAMA_KV_OVERRIDE_TYPE_STR;
-        kvo.str_value = strdup(sep);
+        strncpy(kvo.str_value, sep, 128);
     } else {
         fprintf(stderr, "%s: invalid type for KV override '%s'\n", __func__, data);
         return false;
@@ -2232,10 +2232,6 @@ 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);
@@ -2956,11 +2952,3 @@ 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 916023e25..562d3a119 100644
--- a/common/common.h
+++ b/common/common.h
@@ -172,9 +172,6 @@ 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/examples/quantize/quantize.cpp b/examples/quantize/quantize.cpp
index a0d9a2f46..b6464be3d 100644
--- a/examples/quantize/quantize.cpp
+++ b/examples/quantize/quantize.cpp
@@ -305,14 +305,14 @@ int main(int argc, char ** argv) {
             llama_model_kv_override kvo;
             std::strcpy(kvo.key, LLM_KV_QUANTIZE_IMATRIX_FILE);
             kvo.tag = LLAMA_KV_OVERRIDE_TYPE_STR;
-            kvo.str_value = strdup(imatrix_file.c_str());
+            strncpy(kvo.str_value, imatrix_file.c_str(), 128);
             kv_overrides.emplace_back(std::move(kvo));
         }
         if (!imatrix_dataset.empty()) {
             llama_model_kv_override kvo;
             std::strcpy(kvo.key, LLM_KV_QUANTIZE_IMATRIX_DATASET);
             kvo.tag = LLAMA_KV_OVERRIDE_TYPE_STR;
-            kvo.str_value = strdup(imatrix_dataset.c_str());
+            strncpy(kvo.str_value, imatrix_dataset.c_str(), 128);
             kv_overrides.emplace_back(std::move(kvo));
         }
 
diff --git a/llama.h b/llama.h
index 0cc0d3f29..73d9733f7 100644
--- a/llama.h
+++ b/llama.h
@@ -201,11 +201,11 @@ extern "C" {
     struct llama_model_kv_override {
         char key[128];
         enum llama_model_kv_override_type tag;
+        char str_value[128];
         union {
             int64_t int_value;
             double  float_value;
             bool    bool_value;
-            char *  str_value;
         };
     };