Daniel Hiltgen c146a138e3 ggml: handle all streams (#13350) ...

Follow up from #12992

Free all streams, and keep the alloc logic aligned across streams.

2025-12-05 16:10:33 -08:00

..

backend

ggml: handle all streams (#13350)

2025-12-05 16:10:33 -08:00

nn

ggml: Enable flash attention for vision encoders

2025-12-04 15:19:06 -08:00

backend.go

ggml: Enable flash attention for vision encoders

2025-12-04 15:19:06 -08:00

device.go

CUDA: filter devices on secondary discovery (#13317)

2025-12-03 12:58:16 -08:00

path.go

cpu: always ensure LibOllamaPath included (#12890)

2025-10-31 14:37:29 -07:00