Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

ggml: Don't allocate CPU buffers as CUDA Host buffers

Jesse Gross34c3b68fc8a1
diff --git a/ml/backend/ggml/ggml.go b/ml/backend/ggml/ggml.go
index 727a7b8e5..b4644d97e 100644
--- a/ml/backend/ggml/ggml.go
+++ b/ml/backend/ggml/ggml.go
@@ -384,12 +384,6 @@ func New(ctx context.Context, r *os.File, params ml.BackendParams) (ml.Backend,
for _, d := range append(gpus, append(accels, cpus...)...) {
b := C.ggml_backend_dev_init(d, nil)
bt := C.ggml_backend_get_default_buffer_type(b)
- if d := C.ggml_backend_get_device(b); C.ggml_backend_dev_type(d) == C.GGML_BACKEND_DEVICE_TYPE_CPU && len(gpus) > 0 {
- // use the first gpu host buffer type for gpu if possible
- if hbt := C.ggml_backend_dev_host_buffer_type(gpus[0]); hbt != nil {
- bt = hbt
- }
- }
deviceBufferTypes[d] = bt