Skip to content

[Chore] Run the API tests on torchnative in CI, with capability-based skips #140

Closed

Description

@b-re-w

Prepare the torchnative verification of #84 (M1) and M2 before torchnative's cb-path build, #30 and TN-M1 land: a CI job that installs --extra torchnative (version as an input, default the current PyPI pre-release) and runs the same api/tests (real-model tests included), plus skip conditions that name the missing torchnative capability (e.g. continuous batching unavailable → batching tests skip with the reason; q8_0 quantisation; mps) instead of failing. On upstream PyTorch nothing may skip.

Done when: the job exists (manual + scheduled, non-blocking until torchnative is ready), the skip conditions are documented in SPEC/AGENTS, and a dry run against torchnative 0.1.0b4 shows which tests skip and why.

Activity

  1. b-re-w commented on Oct 3, 2026

    @b-re-w
    MemberAuthor

    Landed in develop via #141. First torchnative run (0.1.0b4, Linux CPU, SmolLM2-135M, run 37103684525): 411 passed, 8 skipped — all 8 are continuous-batching tests, skipped because transformers' CB allocates with pin_memory, which torchnative 0.1.0b4 does not implement (torchnative #30).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions