summaryrefslogtreecommitdiff
path: root/llama.h
diff options
context:
space:
mode:
Diffstat (limited to 'llama.h')
-rw-r--r--llama.h5
1 files changed, 4 insertions, 1 deletions
diff --git a/llama.h b/llama.h
index 7b3634aa..d11cba81 100644
--- a/llama.h
+++ b/llama.h
@@ -6,6 +6,9 @@
#ifdef GGML_USE_CUBLAS
#include "ggml-cuda.h"
#define LLAMA_MAX_DEVICES GGML_CUDA_MAX_DEVICES
+#elif defined(GGML_USE_SYCL)
+#include "ggml-sycl.h"
+#define LLAMA_MAX_DEVICES GGML_SYCL_MAX_DEVICES
#else
#define LLAMA_MAX_DEVICES 1
#endif // GGML_USE_CUBLAS
@@ -46,7 +49,7 @@
#define LLAMA_SESSION_MAGIC LLAMA_FILE_MAGIC_GGSN
#define LLAMA_SESSION_VERSION 4
-#if defined(GGML_USE_CUBLAS) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_METAL)
+#if defined(GGML_USE_CUBLAS) || defined(GGML_USE_CLBLAST) || defined(GGML_USE_METAL) || defined(GGML_USE_SYCL)
// Defined when llama.cpp is compiled with support for offloading model layers to GPU.
#define LLAMA_SUPPORTS_GPU_OFFLOAD
#endif