Repository navigation
chore(deps): pin flopscope 0.12.1 and whestbench 0.16.1 - #40
Conversation
flopscope 0.12.1 makes a symmetry tag describe the buffer it is attached to: as_symmetric validates within a tolerance and now makes the data match the claim before tagging. Data that is already exactly symmetric is untouched and costs the same, which covers nearly every use in the kit. The one change that moves a FLOP figure is full/full_like with a non-scalar fill_value, which no longer infers a symmetry from shape. Announced in the changelog, since the kit is the published spec of the live round. Verified against the published wheels: 47 tests pass, including the cost-docs and rounds-doc suites that assert documented figures against real behaviour.
The kit described the evaluator as sitting on whestbench v0.15.0 with flopscope[server]==0.11.0, and pointed at an open PR to move it. The evaluator has been on v0.16.0 + 0.12.0 for some time, so the claim was two releases behind in four places: the pyproject pin note, the changelog entry for the 0.16.0 move, the cost-docs test header, and the FAQ's atomic-pair answer. With this bump the kit leads the grader by one patch on each, which the pin note now states, along with why that is safe: the only price 0.12.1 moves is full/full_like given a non-scalar fill_value, and nothing in docs/ builds an array that way.
|
Folded in a second commit correcting stale grader-pin claims found while checking the changelog for stale values. The kit described the evaluator as still on With this bump the kit leads the grader by one patch on each rather than trailing it. The pin note now says so, and says why it is safe: the only price 0.12.1 moves is |
Raises the pinned floors to
flopscope>=0.12.1,<0.13.0andwhestbench>=0.16.1,<0.17.0, matching the releases that just shipped.What changed in the meter. flopscope 0.12.1 makes a symmetry tag describe the buffer it is attached to.
flops.as_symmetric()validates within a tolerance, and it now makes the data exactly match the claim before tagging — each orbit takes its first entry. Data that is already exactly symmetric, such as an identity or a covariance built asX.T @ X, is untouched and costs the same; that covers nearly every use in the kit.flops.symmetrize()gainsmode="canonical-copy"beside the unchanged default, and it remains available for symmetry groups too large for the default to enumerate.What moves a FLOP figure.
fnp.fullandfnp.full_likeno longer infer a symmetry from shape whenfill_valueis an array rather than a scalar, so results built that way are priced as the dense arrays they are. A scalar fill is unaffected at every shape. Announced in the changelog, since the kit is the published spec of the live round.Verified against the published wheels (flopscope 0.12.1, whestbench 0.16.1): 47 tests pass, including
test_flopscope_cost_docs.pyandtest_rounds_doc.py, which assert documented figures against real behaviour rather than restating them.