Merge pull request #9824 from ollama/mxyng/sched

conditionally enable parallel pipelines

Merge pull request #9824 from ollama/mxyng/sched
conditionally enable parallel pipelines
021dcf08 · Michael Yang · GitHub · bf24498b · 4561fff3 · 021dcf08
Unverified Commit 021dcf08 authored Mar 17, 2025 by Michael Yang Committed by GitHub Mar 17, 2025
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

ml/backend/ggml/ggml.go ml/backend/ggml/ggml.go +1 -1

No files found.
--- a/ml/backend/ggml/ggml.go
+++ b/ml/backend/ggml/ggml.go
@@ -373,7 +373,7 @@ func New(r *os.File, params ml.BackendParams) (ml.Backend, error) {
 			(*C.ggml_backend_buffer_type_t)(unsafe.Pointer(&schedBufts[0])),
 			C.int(len(schedBackends)),
 			C.size_t(maxGraphNodes),
-			true,
+			C._Bool(len(gpus) > 1 && slices.Contains(gpus, output.d)),
 		),
 		input:  deviceBufferTypes[input.d],
 		output: deviceBufferTypes[output.d],