llama : add missing call to ggml_backend_load_all() (#22752)

Signed-off-by: Adrien Gallouët <redacted>

diff --git a/src/llama.cpp b/src/llama.cpp
index b5dd8433c4b1998e1d19ad2de1ec66f3966ee10f..dfe30ce8f611bad8e7b1ad7585e750d8c6b91d81 100644 (file)
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -71,12 +71,18 @@ bool llama_supports_mlock(void) {
 }
 
 bool llama_supports_gpu_offload(void) {
+    if (!ggml_backend_reg_count()) {
+        ggml_backend_load_all();
+    }
     return ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_GPU) != nullptr ||
            ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_IGPU) != nullptr ||
            llama_supports_rpc();
 }
 
 bool llama_supports_rpc(void) {
+    if (!ggml_backend_reg_count()) {
+        ggml_backend_load_all();
+    }
     return ggml_backend_reg_by_name("RPC") != nullptr;
 }