Egothor/Radixor
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Refresh multilingual benchmarks and fix overlapping gold evaluation

Browse Source 
Recompute published benchmark results for all default language models,
exclude Polish Polimorf, add Hebrew documentation, and record the current
benchmark environment. Evaluate repeated surface forms as an overlapping
gold cover and publish only applicable metrics for candidate policies.
This commit is contained in:
Leo Galambos
2026-07-23 17:06:41 +02:00
parent 1f1b03c6a8
commit b29699b763
64 changed files with 5392 additions and 4765 deletions
Download Patch File Download Diff File Expand all files Collapse all files

115
docs/assets/stylesheets/extra.css
View File

@@ -67,64 +67,62 @@
padding: 0.45rem 0.7rem;
}
/* Publication-quality benchmark tables retain identity columns while scrolling. */
.quality-table {
max-width: 100%;
overflow-x: auto;
/* Primary quality rankings fit the content column; raw counts remain in details. */
.quality-summary {
width: 100%;
margin: 0.65rem 0 1rem;
border: 1px solid var(--md-default-fg-color--lightest);
border-radius: 0.2rem;
scrollbar-gutter: stable;
font-size: 0.72rem;
}
.quality-table:focus {
outline: 0.15rem solid var(--md-accent-fg-color);
outline-offset: 0.1rem;
}
.quality-table::before {
content: "Scrollable table: Rank, Stemmer, and Output policy remain visible.";
.quality-summary .md-typeset__table,
.quality-summary table {
display: block;
padding: 0.35rem 0.55rem;
color: var(--md-default-fg-color--light);
font-size: 0.68rem;
}
.quality-table .md-typeset__table,
.quality-table table {
width: 100%;
margin: 0;
overflow: visible;
}
.quality-table table th:nth-child(1),
.quality-table table td:nth-child(1),
.quality-table table th:nth-child(2),
.quality-table table td:nth-child(2),
.quality-table table th:nth-child(3),
.quality-table table td:nth-child(3) {
position: sticky;
z-index: 2;
background: var(--md-default-bg-color);
background-clip: padding-box;
.quality-summary table {
display: table;
table-layout: fixed;
}
.quality-table table th:nth-child(1),
.quality-table table td:nth-child(1) {
left: 0;
min-width: 2.8rem;
.quality-summary th,
.quality-summary td {
padding: 0.4rem 0.45rem !important;
line-height: 1.3;
letter-spacing: 0;
}
.quality-table table th:nth-child(2),
.quality-table table td:nth-child(2) {
left: 2.8rem;
min-width: 13rem;
white-space: normal;
.quality-summary th:nth-child(1),
.quality-summary td:nth-child(1) {
width: 7%;
}
.quality-table table th:nth-child(3),
.quality-table table td:nth-child(3) {
left: 15.8rem;
min-width: 8.5rem;
box-shadow: 0.2rem 0 0.25rem rgb(0 0 0 / 8%);
.quality-summary th:nth-child(2),
.quality-summary td:nth-child(2) {
width: 39%;
overflow-wrap: anywhere;
}
.quality-summary th:nth-child(n + 3),
.quality-summary td:nth-child(n + 3) {
width: 18%;
}
.quality-summary td:nth-child(n + 3) {
white-space: nowrap;
}
.quality-summary--oracle th:nth-child(1),
.quality-summary--oracle td:nth-child(1) {
width: 46%;
overflow-wrap: anywhere;
}
.quality-summary--oracle th:nth-child(n + 2),
.quality-summary--oracle td:nth-child(n + 2) {
width: 27%;
}
.quality-details > summary {
@@ -132,34 +130,17 @@
}
@media screen and (max-width: 44.99em) {
.quality-table table th:nth-child(2),
.quality-table table td:nth-child(2) {
min-width: 10rem;
.quality-summary {
font-size: 0.62rem;
}
.quality-table table th:nth-child(3),
.quality-table table td:nth-child(3) {
position: static;
min-width: 7.5rem;
box-shadow: none;
.quality-summary th,
.quality-summary td {
padding: 0.3rem 0.2rem !important;
}
}
@media print {
.quality-table {
overflow: visible;
border: 0;
}
.quality-table::before {
display: none;
}
.quality-table table th,
.quality-table table td {
position: static !important;
}
.quality-details:not([open]) > *:not(summary) {
display: block;
}