Skip to content

Labels

Labels

85 labels

  • area:attention

    area:attention
    Exact and experimental attention implementations (attention/)

  • area:ci

    area:ci
    GitHub Actions and repository automation (.github/)

  • area:docs

    area:docs
    README, CONTRIBUTING, BENCHMARKS and other docs

  • area:eval

    area:eval
    Protected evaluator, metrics, artifacts, and policy (eval/, benchmarks/)

  • area:examples

    area:examples
    Runnable examples (examples/)

  • area:matmul

    area:matmul
    Exact engine and tile schedules (matmul/)

  • area:strategy

    area:strategy
    Smart strategies / transforms (strategy/)

  • area:tests

    area:tests
    Correctness gates and test suites (tests/)

  • bug

    bug
    Something isn't working

  • cco-track

    cco-track
    Standing per-track competition (perpetual issue; never closes)

  • cco-winner-dsa_forward

    cco-winner-dsa_forward
    Current dsa_forward champion (king-of-the-hill; earns while held)

  • cco-winner-matmul

    cco-winner-matmul
    Current matmul champion (king-of-the-hill; earns while held)

  • cco-winner-qkv_part_rope

    cco-winner-qkv_part_rope
    Current qkv_part_rope champion (king-of-the-hill; earns while held)

  • cco-winner-rms_norm

    cco-winner-rms_norm
    Current rms_norm champion (king-of-the-hill; earns while held)

  • cco-winner-swiglu_input_quant

    cco-winner-swiglu_input_quant
    Current swiglu_input_quant champion (king-of-the-hill; earns while held)

  • copycat

    copycat
    Blocked: substantially reproduces an earlier PR (see .github/COPYCATS.md)

  • copycat-warn

    copycat-warn
    Flagged for maintainer review: significant overlap with an earlier PR

  • discussion

    discussion
    Open-ended; needs design conversation before code.

  • documentation

    documentation
    Improvements or additions to documentation

  • duplicate

    duplicate
    This issue or pull request already exists

  • enhancement

    enhancement
    New feature or request

  • eval:attention-draft

    eval:attention-draft
    Calibration-only attention result; cannot authorize merge or reward

  • eval:BASELINE

    eval:BASELINE
    Legacy square-track baseline label

  • eval:L

    eval:L
    Verified current-benchmark improvement, calibrated large tier

  • eval:M

    eval:M
    Verified current-benchmark improvement, calibrated medium tier

  • eval:none

    eval:none
    Correct, but does not dominate exact on every cost axis (or gain too small to count)

  • eval:REJECT

    eval:REJECT
    Failed correctness, protected regression, or active benchmark requirements

  • eval:S

    eval:S
    Verified current-benchmark improvement, calibrated small tier

  • eval:XL

    eval:XL
    Verified frontier improvement, largest tier (>=50% over the reference anchor)

  • eval:XS

    eval:XS
    Verified frontier improvement, smallest tier (>=2%, the significance floor)