Skip to main content
Premove ITN v0.1.0 has a conservative compatibility target. A platform is listed as validated only after a release wheel is built, installed without a Rust compiler in a clean environment, and tested by the retained Stage 7 workflow.

​
Certification target

Real model certification uses Python 3.11 once per device family. It runs the exact frozen 1,500 inputs and requires every output string to equal the retained Premove ITN release prediction. This is an execution-equivalence check. It is not an accuracy evaluation and it must not be used to tune the model. The workflow records the OS, architecture, Python, PyTorch, Transformers, Rust target, wheel filename and digest, source commit, model revision, reference prediction digest, selected device, result, and observed peak process RSS. The validated table above is based on successful retained workflow evidence. The MPS evidence was produced with the same clean release wheel on an Apple Silicon host because the standard public macOS runner does not have enough shared memory for this model, and the repository does not have access to the larger macOS runner tier.

​
Not validated for v0.1.0

  • Windows
  • macOS Intel
  • Linux ARM64
  • CUDA
  • ROCm and other accelerator backends
The API exposes device="cuda", but CUDA is not a v0.1.0 compatibility claim until it is tested on real NVIDIA hardware. Unsupported or unavailable explicit devices fail with a clear runtime error.