leva 0.2.0 → 0.2.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/README.md +1 -1
- data/app/assets/stylesheets/leva/application.css +156 -25
- data/app/views/leva/workbench/_results_section.html.erb +2 -2
- data/lib/leva/version.rb +1 -1
- metadata +1 -1
checksums.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
---
|
|
2
2
|
SHA256:
|
|
3
|
-
metadata.gz:
|
|
4
|
-
data.tar.gz:
|
|
3
|
+
metadata.gz: 91394838ea4f4d47fd0968806beee6ce6d30ecc11888565630d76ca0a6d48fc0
|
|
4
|
+
data.tar.gz: 48f3178cf8be1f4fafaf9e485b5e6bab93c54570f71d738623e19d15cb6b2a13
|
|
5
5
|
SHA512:
|
|
6
|
-
metadata.gz:
|
|
7
|
-
data.tar.gz:
|
|
6
|
+
metadata.gz: e3f8f96aef6161234ceff59cfb81f18d650b4e617aafcbd53badca722720fd065b206a10b3319bd6569c07d1d71ae50398b35081a526c242f691239421200262
|
|
7
|
+
data.tar.gz: 6f33990542e0a0d4c64611fdd5bab3ff6af7b42838350a414801a997ce5aefc72ce54f7c40ee6b1977fd6523616fc18c1d4a036ff02f44e82424b74acf4ad8e9
|
data/README.md
CHANGED
|
@@ -5,7 +5,7 @@
|
|
|
5
5
|
|
|
6
6
|
Leva is a Ruby on Rails framework for evaluating Language Models (LLMs) using ActiveRecord datasets on production models. It provides a flexible structure for creating experiments, managing datasets, and implementing various evaluation logic on production data with security in mind.
|
|
7
7
|
|
|
8
|
-
|
|
9
9
|
|
|
10
10
|
## Installation
|
|
11
11
|
|
|
@@ -1330,6 +1330,13 @@ strong {
|
|
|
1330
1330
|
padding: var(--space-4);
|
|
1331
1331
|
overflow-y: auto;
|
|
1332
1332
|
transition: width var(--transition-base), padding var(--transition-base);
|
|
1333
|
+
/* Hide scrollbar while maintaining scroll */
|
|
1334
|
+
scrollbar-width: none;
|
|
1335
|
+
-ms-overflow-style: none;
|
|
1336
|
+
}
|
|
1337
|
+
|
|
1338
|
+
.panel-right::-webkit-scrollbar {
|
|
1339
|
+
display: none;
|
|
1333
1340
|
}
|
|
1334
1341
|
|
|
1335
1342
|
.panel-right.resizing {
|
|
@@ -1833,6 +1840,40 @@ dialog::backdrop {
|
|
|
1833
1840
|
}
|
|
1834
1841
|
|
|
1835
1842
|
|
|
1843
|
+
/* ============================================
|
|
1844
|
+
SCROLLBAR HIDING (while maintaining scroll)
|
|
1845
|
+
============================================ */
|
|
1846
|
+
|
|
1847
|
+
/* Hide scrollbars globally in workbench panels */
|
|
1848
|
+
.panel,
|
|
1849
|
+
.panel-right,
|
|
1850
|
+
.sidebar-content,
|
|
1851
|
+
.output-block,
|
|
1852
|
+
.result-block,
|
|
1853
|
+
.main-content,
|
|
1854
|
+
.prompt-textarea,
|
|
1855
|
+
.prompt-sidebar,
|
|
1856
|
+
.layout-workbench,
|
|
1857
|
+
.content-body,
|
|
1858
|
+
.output-value {
|
|
1859
|
+
scrollbar-width: none; /* Firefox */
|
|
1860
|
+
-ms-overflow-style: none; /* IE 10+ */
|
|
1861
|
+
}
|
|
1862
|
+
|
|
1863
|
+
.panel::-webkit-scrollbar,
|
|
1864
|
+
.panel-right::-webkit-scrollbar,
|
|
1865
|
+
.sidebar-content::-webkit-scrollbar,
|
|
1866
|
+
.output-block::-webkit-scrollbar,
|
|
1867
|
+
.result-block::-webkit-scrollbar,
|
|
1868
|
+
.main-content::-webkit-scrollbar,
|
|
1869
|
+
.prompt-textarea::-webkit-scrollbar,
|
|
1870
|
+
.prompt-sidebar::-webkit-scrollbar,
|
|
1871
|
+
.layout-workbench::-webkit-scrollbar,
|
|
1872
|
+
.content-body::-webkit-scrollbar,
|
|
1873
|
+
.output-value::-webkit-scrollbar {
|
|
1874
|
+
display: none; /* Chrome, Safari, Opera */
|
|
1875
|
+
}
|
|
1876
|
+
|
|
1836
1877
|
/* ============================================
|
|
1837
1878
|
WORKBENCH SPECIFIC
|
|
1838
1879
|
============================================ */
|
|
@@ -1993,9 +2034,11 @@ dialog::backdrop {
|
|
|
1993
2034
|
============================================ */
|
|
1994
2035
|
|
|
1995
2036
|
.run-controls {
|
|
1996
|
-
padding
|
|
1997
|
-
|
|
1998
|
-
|
|
2037
|
+
padding: var(--space-4);
|
|
2038
|
+
background: rgba(26, 25, 24, 0.4);
|
|
2039
|
+
border-radius: var(--radius-lg);
|
|
2040
|
+
border: 1px solid var(--gray-800);
|
|
2041
|
+
margin-bottom: var(--space-3);
|
|
1999
2042
|
}
|
|
2000
2043
|
|
|
2001
2044
|
.run-selects {
|
|
@@ -2024,22 +2067,26 @@ dialog::backdrop {
|
|
|
2024
2067
|
============================================ */
|
|
2025
2068
|
|
|
2026
2069
|
.output-section {
|
|
2027
|
-
padding
|
|
2028
|
-
|
|
2029
|
-
|
|
2070
|
+
padding: var(--space-4);
|
|
2071
|
+
background: rgba(26, 25, 24, 0.4);
|
|
2072
|
+
border-radius: var(--radius-lg);
|
|
2073
|
+
margin-bottom: var(--space-3);
|
|
2074
|
+
border: 1px solid var(--gray-800);
|
|
2030
2075
|
}
|
|
2031
2076
|
|
|
2032
2077
|
.output-header {
|
|
2033
2078
|
display: flex;
|
|
2034
|
-
align-items:
|
|
2079
|
+
align-items: center;
|
|
2035
2080
|
justify-content: space-between;
|
|
2036
|
-
margin-bottom: var(--space-
|
|
2081
|
+
margin-bottom: var(--space-4);
|
|
2082
|
+
padding-bottom: var(--space-3);
|
|
2083
|
+
border-bottom: 1px solid var(--gray-800);
|
|
2037
2084
|
}
|
|
2038
2085
|
|
|
2039
2086
|
.output-title {
|
|
2040
2087
|
font-size: var(--text-xs);
|
|
2041
2088
|
font-weight: 600;
|
|
2042
|
-
color: var(--gray-
|
|
2089
|
+
color: var(--gray-300);
|
|
2043
2090
|
text-transform: uppercase;
|
|
2044
2091
|
letter-spacing: var(--tracking-wide);
|
|
2045
2092
|
}
|
|
@@ -2047,48 +2094,125 @@ dialog::backdrop {
|
|
|
2047
2094
|
.output-meta {
|
|
2048
2095
|
font-size: 10px;
|
|
2049
2096
|
color: var(--gray-500);
|
|
2097
|
+
display: flex;
|
|
2098
|
+
align-items: center;
|
|
2099
|
+
gap: var(--space-2);
|
|
2050
2100
|
}
|
|
2051
2101
|
|
|
2052
2102
|
.output-grid {
|
|
2053
2103
|
display: flex;
|
|
2054
2104
|
flex-direction: column;
|
|
2055
|
-
gap: var(--space-
|
|
2105
|
+
gap: var(--space-3);
|
|
2056
2106
|
}
|
|
2057
2107
|
|
|
2058
2108
|
.output-block {
|
|
2059
2109
|
background: var(--gray-800);
|
|
2060
|
-
border-radius: var(--radius-
|
|
2061
|
-
padding: var(--space-
|
|
2110
|
+
border-radius: var(--radius-md);
|
|
2111
|
+
padding: var(--space-3) var(--space-4);
|
|
2062
2112
|
display: flex;
|
|
2063
|
-
align-items:
|
|
2113
|
+
align-items: flex-start;
|
|
2064
2114
|
gap: var(--space-3);
|
|
2115
|
+
transition: background-color 0.15s ease, border-color 0.15s ease;
|
|
2116
|
+
}
|
|
2117
|
+
|
|
2118
|
+
.output-block:hover {
|
|
2119
|
+
background: var(--gray-750, #2a2928);
|
|
2065
2120
|
}
|
|
2066
2121
|
|
|
2067
2122
|
.output-block--expected {
|
|
2068
|
-
background: rgba(125, 179, 103, 0.
|
|
2069
|
-
border-left:
|
|
2123
|
+
background: rgba(125, 179, 103, 0.06);
|
|
2124
|
+
border-left: 3px solid var(--success-500);
|
|
2125
|
+
padding: var(--space-2) var(--space-3);
|
|
2126
|
+
}
|
|
2127
|
+
|
|
2128
|
+
.output-block--expected:hover {
|
|
2129
|
+
background: rgba(125, 179, 103, 0.1);
|
|
2130
|
+
}
|
|
2131
|
+
|
|
2132
|
+
.output-block--expected .output-value {
|
|
2133
|
+
font-weight: 500;
|
|
2134
|
+
color: var(--gray-50);
|
|
2070
2135
|
}
|
|
2071
2136
|
|
|
2072
2137
|
.output-block--got {
|
|
2073
|
-
|
|
2138
|
+
background: var(--gray-850, #222120);
|
|
2139
|
+
border: 1px solid var(--gray-700);
|
|
2140
|
+
border-left: 3px solid var(--gray-600);
|
|
2141
|
+
}
|
|
2142
|
+
|
|
2143
|
+
.output-block--parsed {
|
|
2144
|
+
background: rgba(212, 168, 74, 0.06);
|
|
2145
|
+
border-left: 3px solid var(--accent-500);
|
|
2146
|
+
padding: var(--space-2) var(--space-3);
|
|
2147
|
+
}
|
|
2148
|
+
|
|
2149
|
+
.output-block--parsed:hover {
|
|
2150
|
+
background: rgba(212, 168, 74, 0.1);
|
|
2151
|
+
}
|
|
2152
|
+
|
|
2153
|
+
.output-block--parsed .output-value {
|
|
2154
|
+
font-weight: 500;
|
|
2155
|
+
color: var(--accent-300);
|
|
2074
2156
|
}
|
|
2075
2157
|
|
|
2076
2158
|
.output-label {
|
|
2077
2159
|
font-size: 10px;
|
|
2078
2160
|
font-weight: 600;
|
|
2079
|
-
color: var(--gray-
|
|
2161
|
+
color: var(--gray-400);
|
|
2080
2162
|
text-transform: uppercase;
|
|
2081
|
-
letter-spacing: 0.
|
|
2163
|
+
letter-spacing: 0.05em;
|
|
2082
2164
|
flex-shrink: 0;
|
|
2083
|
-
width:
|
|
2165
|
+
min-width: 64px;
|
|
2166
|
+
padding-top: 2px;
|
|
2084
2167
|
}
|
|
2085
2168
|
|
|
2086
2169
|
.output-value {
|
|
2087
2170
|
font-family: var(--font-mono);
|
|
2088
2171
|
font-size: var(--text-sm);
|
|
2089
2172
|
color: var(--gray-100);
|
|
2090
|
-
line-height: 1.
|
|
2173
|
+
line-height: 1.6;
|
|
2091
2174
|
word-break: break-word;
|
|
2175
|
+
white-space: pre-wrap;
|
|
2176
|
+
flex: 1;
|
|
2177
|
+
min-width: 0;
|
|
2178
|
+
}
|
|
2179
|
+
|
|
2180
|
+
/* Long output values get constrained height with scroll */
|
|
2181
|
+
.output-block--got .output-value {
|
|
2182
|
+
max-height: 180px;
|
|
2183
|
+
overflow-y: auto;
|
|
2184
|
+
scrollbar-width: none;
|
|
2185
|
+
-ms-overflow-style: none;
|
|
2186
|
+
font-size: 12px;
|
|
2187
|
+
line-height: 1.7;
|
|
2188
|
+
color: var(--gray-200);
|
|
2189
|
+
}
|
|
2190
|
+
|
|
2191
|
+
.output-block--got .output-value::-webkit-scrollbar {
|
|
2192
|
+
display: none;
|
|
2193
|
+
}
|
|
2194
|
+
|
|
2195
|
+
/* Fade effect for long content */
|
|
2196
|
+
.output-block--got {
|
|
2197
|
+
position: relative;
|
|
2198
|
+
}
|
|
2199
|
+
|
|
2200
|
+
.output-block--got::after {
|
|
2201
|
+
content: '';
|
|
2202
|
+
position: absolute;
|
|
2203
|
+
bottom: 0;
|
|
2204
|
+
left: 0;
|
|
2205
|
+
right: 0;
|
|
2206
|
+
height: 40px;
|
|
2207
|
+
background: linear-gradient(transparent, var(--gray-850, #222120));
|
|
2208
|
+
pointer-events: none;
|
|
2209
|
+
border-radius: 0 0 var(--radius-md) var(--radius-md);
|
|
2210
|
+
opacity: 0;
|
|
2211
|
+
transition: opacity 0.2s;
|
|
2212
|
+
}
|
|
2213
|
+
|
|
2214
|
+
.output-block--got:has(.output-value:not(:hover)):after {
|
|
2215
|
+
opacity: 0.8;
|
|
2092
2216
|
}
|
|
2093
2217
|
|
|
2094
2218
|
.output-empty {
|
|
@@ -2103,20 +2227,25 @@ dialog::backdrop {
|
|
|
2103
2227
|
============================================ */
|
|
2104
2228
|
|
|
2105
2229
|
.eval-section {
|
|
2106
|
-
|
|
2230
|
+
padding: var(--space-4);
|
|
2231
|
+
background: rgba(26, 25, 24, 0.4);
|
|
2232
|
+
border-radius: var(--radius-lg);
|
|
2233
|
+
border: 1px solid var(--gray-800);
|
|
2107
2234
|
}
|
|
2108
2235
|
|
|
2109
2236
|
.eval-header {
|
|
2110
2237
|
display: flex;
|
|
2111
2238
|
align-items: center;
|
|
2112
2239
|
justify-content: space-between;
|
|
2113
|
-
margin-bottom: var(--space-
|
|
2240
|
+
margin-bottom: var(--space-4);
|
|
2241
|
+
padding-bottom: var(--space-3);
|
|
2242
|
+
border-bottom: 1px solid var(--gray-800);
|
|
2114
2243
|
}
|
|
2115
2244
|
|
|
2116
2245
|
.eval-title {
|
|
2117
2246
|
font-size: var(--text-xs);
|
|
2118
2247
|
font-weight: 600;
|
|
2119
|
-
color: var(--gray-
|
|
2248
|
+
color: var(--gray-300);
|
|
2120
2249
|
text-transform: uppercase;
|
|
2121
2250
|
letter-spacing: var(--tracking-wide);
|
|
2122
2251
|
}
|
|
@@ -2130,8 +2259,8 @@ dialog::backdrop {
|
|
|
2130
2259
|
.eval-card {
|
|
2131
2260
|
flex: 1 1 calc(50% - var(--space-1));
|
|
2132
2261
|
min-width: 0;
|
|
2133
|
-
background: var(--gray-
|
|
2134
|
-
border: 1px solid
|
|
2262
|
+
background: var(--gray-850, #222120);
|
|
2263
|
+
border: 1px solid var(--gray-700);
|
|
2135
2264
|
border-radius: var(--radius-md);
|
|
2136
2265
|
padding: var(--space-3);
|
|
2137
2266
|
cursor: pointer;
|
|
@@ -2143,6 +2272,8 @@ dialog::backdrop {
|
|
|
2143
2272
|
|
|
2144
2273
|
.eval-card:hover {
|
|
2145
2274
|
background: var(--gray-750, #2a2928);
|
|
2275
|
+
border-color: var(--gray-600);
|
|
2276
|
+
transform: translateY(-1px);
|
|
2146
2277
|
}
|
|
2147
2278
|
|
|
2148
2279
|
.eval-card-inner {
|
|
@@ -83,9 +83,9 @@
|
|
|
83
83
|
<span class="output-value"><%= runner_result.prediction %></span>
|
|
84
84
|
</div>
|
|
85
85
|
<% if runner_result.dataset_record.recordable.extract_regex_pattern && runner_result.parsed_predictions.any? %>
|
|
86
|
-
<div class="output-block
|
|
86
|
+
<div class="output-block output-block--parsed">
|
|
87
87
|
<span class="output-label">
|
|
88
|
-
<svg class="icon-sm" style="display: inline; width: 12px; height: 12px; margin-right: 4px; vertical-align: -2px; color: var(--
|
|
88
|
+
<svg class="icon-sm" style="display: inline; width: 12px; height: 12px; margin-right: 4px; vertical-align: -2px; color: var(--accent-500);" fill="none" viewBox="0 0 24 24" stroke="currentColor">
|
|
89
89
|
<path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M7 21a4 4 0 01-4-4V5a2 2 0 012-2h4a2 2 0 012 2v12a4 4 0 01-4 4zm0 0h12a2 2 0 002-2v-4a2 2 0 00-2-2h-2.343M11 7.343l1.657-1.657a2 2 0 012.828 0l2.829 2.829a2 2 0 010 2.828l-8.486 8.485M7 17h.01" />
|
|
90
90
|
</svg>
|
|
91
91
|
Parsed
|
data/lib/leva/version.rb
CHANGED