Load model, pledge+unveil.

3 files changed, 34 insertions, 0 deletions

diff --git a/.gitignore b/.gitignore
index 43e4558..926cbed 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,3 +1,4 @@
 *.o
 *.cgi
+*.gguf
 deps/
diff --git a/README.txt b/README.txt
new file mode 100644
index 0000000..c45e800
--- /dev/null
+++ b/README.txt
@@ -0,0 +1,2 @@
+Run with no logs: GGML_LOG_LEVEL=0 ./wordbook.cgi
+Run with error logs: GGML_LOG_LEVEL=error ./wordbook.cgi
diff --git a/main.c b/main.c
index e221a28..c5ecd8a 100644
--- a/main.c
+++ b/main.c
@@ -1,10 +1,41 @@
 #include <stdio.h>
+#include <stdbool.h>
+#include <err.h>
+#include <unistd.h>
+
 #include "llama.h"
 
+#define MODEL_PATH "Qwen2.5-7B-Instruct-Q4_K_M.gguf"
+
 int main(int argc , char *argv[])
 {
+	struct llama_model *model;
+	struct llama_model_params mparams;
+
+	if (unveil(MODEL_PATH, "r") == -1)
+		err(1, "unveil %s failed", MODEL_PATH);
+
+	if (unveil(NULL, NULL) == -1)
+		err(1, "unveil lock failed");
+
+	if (pledge("stdio rpath", NULL) == -1)
+		err(1, "initial pledge failed");
+
 	llama_backend_init();
 
+	mparams = llama_model_default_params();
+	mparams.n_gpu_layers = 0; /* force all layers onto CPU */
+	mparams.load_mode = LLAMA_LOAD_MODE_MMAP;
+
+	model = llama_model_load_from_file(MODEL_PATH, mparams);
+	if (!model)
+		errx(1, "failed to load model from file %s", MODEL_PATH);
+
+	/* drop rpath after loading the model */
+	if (pledge("stdio", NULL) == -1)
+		err(1, "secondary pledge failed");
+
+	llama_model_free(model);
 	llama_backend_free();
 
 	return 0;