Skip to content

fitctl v0.3.0

Choose a tag to compare

@tznurmin tznurmin released this 25 Apr 09:01

fitctl 0.3.0 expands accelerator and runtime-state coverage, adds CUDA-aware validation, and improves operator inspection output.

Added

  • CUDA runtime state collection through fitctl.runtime.cuda
  • CUDA driver, toolkit, driver-supported CUDA, and runtime version reporting
  • Per-device CUDA allocatable, total, and used memory reporting
  • Runtime-aware GPU validation using state
  • Multi-GPU validation with policy-scoped accelerator counts
  • Qualifying-device CUDA memory checks, including per-device floors and aggregate memory over qualifying devices
  • inspect --view coverage for grouped field coverage on survey, contract, and state artifacts
  • machine-id-first local stable identity for stateful host correlation

Changed

  • Accelerator inspect output now prioritizes GPU model, driver, operability, and runtime detail
  • Build provenance fields are kept separate from the release-line fitctl_version