"""Bound local inference workers so foreground input has scheduling headroom. Keep the policy aligned with src/main/utils/inference-cpu-budget.ts. This limits inference workers, not process affinity or total process CPU usage. """ import os def inference_cpu_threads(available_cpus: int | None) -> int: if available_cpus is None or available_cpus < 1: return 1 return max(1, min(4, available_cpus - 2)) def local_inference_cpu_threads() -> int: # Python 3.13+ exposes the CPUs available to this process, including affinity. count = getattr(os, "process_cpu_count", os.cpu_count)() return inference_cpu_threads(count)