whisper.cpp

0b1962a1 - Add `--no-op-offload` to improve `-ot` pp perf in MoE models like llama4 400B (llama/13386)