diff options
| author | Johannes Gäßler <johannesg@5d6.de> | 2023-08-22 22:47:05 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-08-22 22:47:05 +0200 |
| commit | c63bb1d16a70c03440671b76954bb767513cead8 (patch) | |
| tree | 5562dd98e6a37a1bfe635bf25114e3bdb5660996 /ggml-cuda.cu | |
| parent | 3b6cfe7c927df178ca3c11643c3ec93e143471c9 (diff) | |
CUDA: use mul_mat_q kernels by default (#2683)
Diffstat (limited to 'ggml-cuda.cu')
| -rw-r--r-- | ggml-cuda.cu | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/ggml-cuda.cu b/ggml-cuda.cu index 4fe378c2..70a950bb 100644 --- a/ggml-cuda.cu +++ b/ggml-cuda.cu @@ -287,7 +287,7 @@ static int g_device_count = -1; static int g_main_device = 0; static int g_compute_capabilities[GGML_CUDA_MAX_DEVICES]; static float g_tensor_split[GGML_CUDA_MAX_DEVICES] = {0}; -static bool g_mul_mat_q = false; +static bool g_mul_mat_q = true; static void * g_scratch_buffer = nullptr; static size_t g_scratch_size = 1024*1024*1024; // 1 GB by default |