ggml-backend : only offload from host buffers (#11120)
This commit is contained in:
parent
a4dd490069
commit
a3d50bc022
1 changed files with 1 additions and 1 deletions
|
@ -761,7 +761,7 @@ static int ggml_backend_sched_backend_id_from_cur(ggml_backend_sched_t sched, st
|
|||
}
|
||||
// skip ROPE since the rope freqs tensor is too small to choose a backend based on it
|
||||
// not an ideal solution
|
||||
if (tensor->op != GGML_OP_ROPE && src->buffer != NULL && src->buffer->usage == GGML_BACKEND_BUFFER_USAGE_WEIGHTS) {
|
||||
if (tensor->op != GGML_OP_ROPE && src->buffer != NULL && src->buffer->usage == GGML_BACKEND_BUFFER_USAGE_WEIGHTS && ggml_backend_buffer_is_host(src->buffer)) {
|
||||
int src_backend_id = ggml_backend_sched_backend_from_buffer(sched, src, tensor);
|
||||
// check if a backend with higher prio wants to offload the op
|
||||
if (src_backend_id == sched->n_backends - 1) {
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue