Commit History
Today
-
add the preprint link and a citation block
Verified signaturea86ed31
Yesterday
-
warn against the standard pipeline; pipeline_tag no longer implies it
Verified signaturef638719 -
Add the risk-coverage result and say plainly that this is not an ablation
Verified signatureTwo things the preprint carries and the card did not. First, thresholding tells a kinder story than the fixed 0.8 cut: in most cells where accuracy improves the area under the risk-coverage curve falls, from 0.235 to 0.078 on strict jargon with five candidates and from 0.452 to 0.418 with eighteen, with slang on all labels at eighteen candidates the exception at 0.354 to 0.395. So the release is easier to threshold on unfamiliar wording while being worse at a fixed cut, and which of those matters depends on whether the caller tunes a threshold. Second, the two releases differ in more than the label augmentation and each was trained once, so none of this is a controlled ablation of it. An attempt to isolate the augmentation by retraining the recipe on other seeds was void, because the mixture shipped to those runs predated the commit that added the augmentation. That is now stated rather than left for a reader to work out. head.safetensors is included again so the usage snippet and th…9e58317 -
Carry the wording result into the advice a caller actually reads
Verified signatureThe naming section repeated the five-candidate table and told the reader to measure in-house jargon, which is still right but no longer the whole finding. It now says what the bench shows: ordinary words, borrowed words and office slang are where training a label in several forms pays, abbreviations and opaque codes are not read better by this release than by the previous one, and a code that works does so because it contains a readable key. It also notes that the confident-error counts in its table are the five-candidate ones and that they rise with eighteen candidates, and it points at fit_temperature.py for callers who bring their own label vocabulary.d563f68 -
Bring the card to the preprint, and ship the scoring head as safetensors
Verified signatureThe wording bench now reports both candidate-set sizes and marks which rows are controls. The Russian word and the English key are the wordings both releases trained on, so those rows measure recall of a trained form and their near-zero gap is what a control should show; the Russian phrase is a training style for v1.3 only, so its gap is expected by construction. Only the held-out rows speak to generalisation, and the reading they support is narrower than before: training a label in several surface forms helps on unfamiliar wordings made of words, and does nothing for abbreviations or opaque codes, where the label string carries little of its meaning. Confidence no longer follows accuracy in the summary either. With eighteen candidates the count of confident errors rises on jargon and on slang, because the augmented release is more confident overall, so a deployment that invents its own label vocabulary is pointed at fit_temperature.py instead. The calibration figures on the comparab…ac890a1
2 days ago
-
Card: links as a list, fix the same-rows sentence, two wording fixes
Verified signature6aa441b -
Fix the six-task comparison: accuracy on all 10,157 rows, withdraw the calibration claim, add v1.1 confidence
Verified signaturec34cc1b -
Fix the six-task comparison: accuracy on all 10,157 rows, withdraw the calibration claim, add v1.1 confidence
Verified signature071ff25 -
link the interactive demo space
Verified signatured3df0d4 -
card for v1.3: measured numbers only, jargon row explained
Verified signaturedfafb0b -
v1.3: option wording no longer decides the answer; v1.1 kept on its branch
Verified signature85dd0db
4 days ago
-
card for v1.1: routing, label wording, thresholds, baselines and limits
Verified signaturec42f4bd -
v1.1: routing, 5-level reviews and reasoning QA; v1 kept on the v1 branch
Verified signature51d749d -
card: state the held-out calibration error, the shipped 0.0054 is in-sample
Verified signaturef90d279 -
card: link the code repository, drop the Space link, add discovery tags
Verified signatureb41ef19 -
card: project page link and measured CPU latency
Verified signature0c432e4 -
card: add v1 vs v2 comparison, drop the next-version promise
Verified signature9c84342
5 days ago
-
Upload folder using huggingface_hub
Verified signaturec368046 -
initial commit
Verified signatureab228fd