Changelog

Full Changelog: https://github.com/farukciftler/tacet-cli/compare/v0.1.26...v0.1.27

What's Changed

Measure the unmeasured: a grammar property test, the three defects it found, and the Linux sandbox CI can no longer skip by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/1

fix(docs): restore the README phrase the skip-count test pins by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/2

fix(cli): a build that cannot run a model no longer answers as if it could by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/3

feat(eval): the selection suite says how far along it is by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/4

perf(grammar): the mask stopped re-deriving a constant 147k times a step by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/5

fix: a tool call could take 13 minutes — the whitespace cycle, the calendar deadline, and the trace that found them by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/6

fix: three ways one runaway got out — the text ceiling and the tool-call cap by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/7

fix(tools,grammar): the shape only Turkish wrote, and the root cause that stayed shut by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/8

Linux and Windows measured on real machines — what CI could not see by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/9

Two measurements from a rented GPU: what --compare never checked, and why parallel eval wins nothing by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/10

The multiplier a bigger GPU can actually give: a batched-decode measurement by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/11

The model the binary downloads was not the model the numbers describe by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/12

--compare read a missing tool as a broken model by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/13

README: six minutes on a GPU, and where the multiplier actually is by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/14

The three failures the model set actually named — and a README that had drifted by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/15

A benchmark protocol, 608 cases in seven languages, and four models measured by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/16

The llama architecture, two task benchmarks, and the web as a cassette by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/17

Distilling a tool-caller: 0% to 61.5% correct calls, and what it cost by @farukciftler in https://github.com/farukciftler/tacet-cli/pull/18

New Contributors

• @farukciftler made their first contribution in https://github.com/farukciftler/tacet-cli/pull/1

Full Changelog: https://github.com/farukciftler/tacet-cli/compare/v0.1.25...v0.1.26

Full Changelog: https://github.com/farukciftler/tacet-cli/compare/v0.1.24...v0.1.25

Full Changelog: https://github.com/farukciftler/tacet-cli/compare/v0.1.23...v0.1.24

Full Changelog: https://github.com/farukciftler/tacet-cli/compare/v0.1.22...v0.1.23