benchkit-cli 0.1.0__py3-none-any.whl
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.
- benchkit/__init__.py +1 -0
- benchkit/_pi_proxy.py +219 -0
- benchkit/answer_key_guard.ts +49 -0
- benchkit/artifacts.py +87 -0
- benchkit/benchmarks/__init__.py +105 -0
- benchkit/benchmarks/aider_polyglot.py +405 -0
- benchkit/benchmarks/arc.py +52 -0
- benchkit/benchmarks/banking77.py +149 -0
- benchkit/benchmarks/base.py +19 -0
- benchkit/benchmarks/boolq.py +49 -0
- benchkit/benchmarks/chartqa.py +133 -0
- benchkit/benchmarks/code_verifier.py +77 -0
- benchkit/benchmarks/downloads.py +166 -0
- benchkit/benchmarks/evalplus.py +288 -0
- benchkit/benchmarks/git_surgery.py +485 -0
- benchkit/benchmarks/gpqa.py +56 -0
- benchkit/benchmarks/gsm8k.py +97 -0
- benchkit/benchmarks/hellaswag.py +49 -0
- benchkit/benchmarks/humaneval.py +103 -0
- benchkit/benchmarks/ifeval.py +56 -0
- benchkit/benchmarks/ifeval_instructions.py +477 -0
- benchkit/benchmarks/mbpp.py +60 -0
- benchkit/benchmarks/mc_arena.py +479 -0
- benchkit/benchmarks/mcq.py +94 -0
- benchkit/benchmarks/medfailbench.py +537 -0
- benchkit/benchmarks/mmlu.py +54 -0
- benchkit/benchmarks/mmlu_pro.py +62 -0
- benchkit/benchmarks/mmvp.py +115 -0
- benchkit/benchmarks/ocrbench.py +96 -0
- benchkit/benchmarks/openbookqa.py +52 -0
- benchkit/benchmarks/patcheval.py +929 -0
- benchkit/benchmarks/piqa.py +52 -0
- benchkit/benchmarks/ruler.py +172 -0
- benchkit/benchmarks/ruler_generator.py +535 -0
- benchkit/benchmarks/sanity.py +174 -0
- benchkit/benchmarks/treejs_arena.py +268 -0
- benchkit/benchmarks/truthfulqa.py +52 -0
- benchkit/benchmarks/utils.py +25 -0
- benchkit/benchmarks/vision.py +230 -0
- benchkit/benchmarks/vstar.py +108 -0
- benchkit/benchmarks/winogrande.py +52 -0
- benchkit/benchmarks/xstest.py +116 -0
- benchkit/browser.py +429 -0
- benchkit/catalog.py +121 -0
- benchkit/cli.py +1000 -0
- benchkit/client.py +1413 -0
- benchkit/controls.py +87 -0
- benchkit/datasets/MMLU_LICENSE.txt +21 -0
- benchkit/datasets/README.md +142 -0
- benchkit/datasets/arc_challenge.jsonl +1172 -0
- benchkit/datasets/banking77.jsonl +3080 -0
- benchkit/datasets/banking77_intents.json +79 -0
- benchkit/datasets/boolq.jsonl +3270 -0
- benchkit/datasets/code_verifier.jsonl +500 -0
- benchkit/datasets/gpqa_diamond.jsonl +198 -0
- benchkit/datasets/gsm8k.jsonl +1319 -0
- benchkit/datasets/hellaswag.jsonl +1000 -0
- benchkit/datasets/humaneval.jsonl +164 -0
- benchkit/datasets/ifeval.jsonl +541 -0
- benchkit/datasets/mbpp.jsonl +500 -0
- benchkit/datasets/mc_blocks_1_20_1.jsonl +1003 -0
- benchkit/datasets/medfailbench_hard30.jsonl +30 -0
- benchkit/datasets/medfailbench_hard30.md +13 -0
- benchkit/datasets/openbookqa.jsonl +500 -0
- benchkit/datasets/piqa.jsonl +1838 -0
- benchkit/datasets/sanity_code.jsonl +6 -0
- benchkit/datasets/truthfulqa.jsonl +817 -0
- benchkit/datasets/winogrande.jsonl +1267 -0
- benchkit/datasets/xstest.csv +451 -0
- benchkit/decision.py +194 -0
- benchkit/demo.py +451 -0
- benchkit/dockerfiles.py +247 -0
- benchkit/engine.py +1270 -0
- benchkit/evaluation.py +195 -0
- benchkit/events.py +141 -0
- benchkit/executor.py +96 -0
- benchkit/git_surgery/backport-release-stack/setup.sh +328 -0
- benchkit/git_surgery/backport-release-stack/verify.sh +119 -0
- benchkit/git_surgery/bisect-the-regression/setup.sh +49 -0
- benchkit/git_surgery/bisect-the-regression/verify.sh +34 -0
- benchkit/git_surgery/rebase-conflict-chain/setup.sh +65 -0
- benchkit/git_surgery/rebase-conflict-chain/verify.sh +43 -0
- benchkit/git_surgery/recover-complex-stash/setup.sh +152 -0
- benchkit/git_surgery/recover-complex-stash/verify.sh +95 -0
- benchkit/git_surgery/recover-lost-work/setup.sh +40 -0
- benchkit/git_surgery/recover-lost-work/verify.sh +36 -0
- benchkit/git_surgery/repair-force-pushed-remote/setup.sh +171 -0
- benchkit/git_surgery/repair-force-pushed-remote/verify.sh +83 -0
- benchkit/git_surgery/revert-merge-with-followups/setup.sh +198 -0
- benchkit/git_surgery/revert-merge-with-followups/verify.sh +98 -0
- benchkit/git_surgery/secret-in-history/README.md +27 -0
- benchkit/git_surgery/secret-in-history/setup.sh +107 -0
- benchkit/git_surgery/secret-in-history/verify.sh +106 -0
- benchkit/git_surgery/split-the-mega-commit/setup.sh +50 -0
- benchkit/git_surgery/split-the-mega-commit/verify.sh +48 -0
- benchkit/git_surgery/untangle-nested-submodules/setup.sh +193 -0
- benchkit/git_surgery/untangle-nested-submodules/verify.sh +90 -0
- benchkit/history.py +474 -0
- benchkit/jobs.py +332 -0
- benchkit/leakage.py +51 -0
- benchkit/looping.py +873 -0
- benchkit/mc_render.py +274 -0
- benchkit/mc_viewer/dist/atlas.png +0 -0
- benchkit/mc_viewer/dist/blockStates.json.gz +0 -0
- benchkit/mc_viewer/dist/version.json +4 -0
- benchkit/mc_viewer/dist/viewer.html +23 -0
- benchkit/mc_viewer/dist/viewer.js +2 -0
- benchkit/mc_viewer/dist/viewer.js.LICENSE.txt +25 -0
- benchkit/mc_viewer/dist/worker.js +2 -0
- benchkit/mc_viewer/dist/worker.js.LICENSE.txt +19 -0
- benchkit/metrics.py +100 -0
- benchkit/ollama.py +60 -0
- benchkit/perf.py +353 -0
- benchkit/perf_report.py +207 -0
- benchkit/perturbations.py +208 -0
- benchkit/pi_agent.py +476 -0
- benchkit/pi_package/package-lock.json +1827 -0
- benchkit/pi_package/package.json +8 -0
- benchkit/report.py +142 -0
- benchkit/report_markdown.py +645 -0
- benchkit/results.py +688 -0
- benchkit/retry.py +163 -0
- benchkit/runner.py +769 -0
- benchkit/sandbox.py +1167 -0
- benchkit/template_check.py +291 -0
- benchkit/templates/arena.html +504 -0
- benchkit/templates/history.html +58 -0
- benchkit/templates/perf.html +509 -0
- benchkit/templates/report.html +769 -0
- benchkit/tui/__init__.py +5 -0
- benchkit/tui/app.py +87 -0
- benchkit/tui/app.tcss +512 -0
- benchkit/tui/formatting.py +84 -0
- benchkit/tui/screens/__init__.py +1 -0
- benchkit/tui/screens/connect.py +191 -0
- benchkit/tui/screens/detail.py +288 -0
- benchkit/tui/screens/help.py +84 -0
- benchkit/tui/screens/modals.py +544 -0
- benchkit/tui/screens/results.py +361 -0
- benchkit/tui/screens/run.py +988 -0
- benchkit/tui/screens/setup.py +733 -0
- benchkit/tui/theme.py +135 -0
- benchkit/tui/widgets.py +120 -0
- benchkit_cli-0.1.0.dist-info/METADATA +294 -0
- benchkit_cli-0.1.0.dist-info/RECORD +149 -0
- benchkit_cli-0.1.0.dist-info/WHEEL +5 -0
- benchkit_cli-0.1.0.dist-info/entry_points.txt +3 -0
- benchkit_cli-0.1.0.dist-info/licenses/LICENSE +201 -0
- benchkit_cli-0.1.0.dist-info/top_level.txt +1 -0
|
@@ -0,0 +1,509 @@
|
|
|
1
|
+
<!doctype html>
|
|
2
|
+
<html lang="en">
|
|
3
|
+
<head>
|
|
4
|
+
<meta charset="utf-8">
|
|
5
|
+
<meta name="viewport" content="width=device-width, initial-scale=1">
|
|
6
|
+
<meta name="color-scheme" content="light dark">
|
|
7
|
+
<title>BenchKit Performance Profile</title>
|
|
8
|
+
<script>
|
|
9
|
+
(() => {
|
|
10
|
+
let stored = null;
|
|
11
|
+
try { stored = localStorage.getItem("benchkit-theme"); } catch (error) { stored = null; }
|
|
12
|
+
document.documentElement.dataset.theme = stored === "dark" ? "dark" : "light";
|
|
13
|
+
})();
|
|
14
|
+
</script>
|
|
15
|
+
<style>
|
|
16
|
+
:root,
|
|
17
|
+
:root[data-theme="light"] {
|
|
18
|
+
color-scheme: light;
|
|
19
|
+
--page: #ffffff;
|
|
20
|
+
--card: #ffffff;
|
|
21
|
+
--ink: #0b0b0b;
|
|
22
|
+
--muted: #6b7280;
|
|
23
|
+
--line: rgba(11, 11, 11, .12);
|
|
24
|
+
--raise: rgba(11, 11, 11, .04);
|
|
25
|
+
--track: rgba(11, 11, 11, .09);
|
|
26
|
+
--green: #059669;
|
|
27
|
+
--green-soft: #10b981;
|
|
28
|
+
--red: #dc2626;
|
|
29
|
+
--blue: #2563eb;
|
|
30
|
+
--amber: #d97706;
|
|
31
|
+
--bar: #6b7280;
|
|
32
|
+
--bar-light: rgba(11, 11, 11, .09);
|
|
33
|
+
--shade-1: #dbeafe;
|
|
34
|
+
--shade-2: #bfdbfe;
|
|
35
|
+
--shade-3: #93c5fd;
|
|
36
|
+
--shade-4: #60a5fa;
|
|
37
|
+
--shade-5: #3b82f6;
|
|
38
|
+
--shade-6: #2563eb;
|
|
39
|
+
--shade-7: #1d4ed8;
|
|
40
|
+
}
|
|
41
|
+
|
|
42
|
+
:root[data-theme="dark"] {
|
|
43
|
+
color-scheme: dark;
|
|
44
|
+
--page: #000000;
|
|
45
|
+
--card: #000000;
|
|
46
|
+
--ink: #ffffff;
|
|
47
|
+
--muted: #6b7280;
|
|
48
|
+
--line: rgba(255, 255, 255, .14);
|
|
49
|
+
--raise: rgba(255, 255, 255, .06);
|
|
50
|
+
--track: rgba(255, 255, 255, .12);
|
|
51
|
+
--green: #34d399;
|
|
52
|
+
--green-soft: #6ee7b7;
|
|
53
|
+
--red: #ef4444;
|
|
54
|
+
--blue: #60a5fa;
|
|
55
|
+
--amber: #facc15;
|
|
56
|
+
--bar: #9aa1ad;
|
|
57
|
+
--bar-light: rgba(255, 255, 255, .12);
|
|
58
|
+
--shade-1: #14304f;
|
|
59
|
+
--shade-2: #1b4470;
|
|
60
|
+
--shade-3: #235891;
|
|
61
|
+
--shade-4: #2c6cb2;
|
|
62
|
+
--shade-5: #3f86d4;
|
|
63
|
+
--shade-6: #4f95e8;
|
|
64
|
+
--shade-7: #60a5fa;
|
|
65
|
+
}
|
|
66
|
+
|
|
67
|
+
:root {
|
|
68
|
+
--sans: "Avenir Next", Avenir, "Century Gothic", sans-serif;
|
|
69
|
+
--mono: "SFMono-Regular", "Cascadia Mono", Menlo, monospace;
|
|
70
|
+
}
|
|
71
|
+
|
|
72
|
+
* { box-sizing: border-box; }
|
|
73
|
+
html { background: var(--page); }
|
|
74
|
+
body {
|
|
75
|
+
margin: 0;
|
|
76
|
+
min-height: 100vh;
|
|
77
|
+
color: var(--ink);
|
|
78
|
+
background: var(--page);
|
|
79
|
+
font-family: var(--sans);
|
|
80
|
+
-webkit-font-smoothing: antialiased;
|
|
81
|
+
}
|
|
82
|
+
button { font: inherit; }
|
|
83
|
+
.page { width: min(1180px, calc(100% - 40px)); margin: 0 auto; padding: 54px 0 70px; }
|
|
84
|
+
|
|
85
|
+
.report-head {
|
|
86
|
+
display: flex;
|
|
87
|
+
justify-content: space-between;
|
|
88
|
+
gap: 32px;
|
|
89
|
+
align-items: end;
|
|
90
|
+
padding-bottom: 30px;
|
|
91
|
+
border-bottom: 1px solid var(--line);
|
|
92
|
+
}
|
|
93
|
+
h1, h2, h3, p { margin-top: 0; }
|
|
94
|
+
h1 { margin-bottom: 7px; font-size: 42px; line-height: 1; letter-spacing: -.045em; }
|
|
95
|
+
.meta { margin: 0; color: var(--muted); font: 11px/1.55 var(--mono); }
|
|
96
|
+
.head-right { display: flex; flex-direction: column; align-items: flex-end; gap: 18px; }
|
|
97
|
+
.best { text-align: right; }
|
|
98
|
+
.best-score { color: var(--green); font-size: 42px; font-weight: 800; line-height: 1; letter-spacing: -.045em; }
|
|
99
|
+
.best-model { margin-top: 6px; font-size: 12px; font-weight: 800; }
|
|
100
|
+
|
|
101
|
+
.theme-toggle {
|
|
102
|
+
display: inline-flex;
|
|
103
|
+
align-items: center;
|
|
104
|
+
gap: 8px;
|
|
105
|
+
padding: 7px 14px;
|
|
106
|
+
border: 1px solid var(--line);
|
|
107
|
+
border-radius: 999px;
|
|
108
|
+
background: var(--raise);
|
|
109
|
+
color: var(--ink);
|
|
110
|
+
cursor: pointer;
|
|
111
|
+
font: 700 9px/1 var(--mono);
|
|
112
|
+
letter-spacing: .09em;
|
|
113
|
+
text-transform: uppercase;
|
|
114
|
+
}
|
|
115
|
+
.theme-toggle:hover { border-color: var(--blue); color: var(--blue); }
|
|
116
|
+
.theme-dot { width: 9px; height: 9px; border-radius: 50%; background: var(--blue); }
|
|
117
|
+
|
|
118
|
+
.summary {
|
|
119
|
+
display: grid;
|
|
120
|
+
grid-template-columns: repeat(4, 1fr);
|
|
121
|
+
margin: 28px 0 0;
|
|
122
|
+
background: var(--card);
|
|
123
|
+
border: 1px solid var(--line);
|
|
124
|
+
}
|
|
125
|
+
.stat { padding: 21px 24px; }
|
|
126
|
+
.stat + .stat { border-left: 1px solid var(--line); }
|
|
127
|
+
.stat-label { color: var(--muted); font: 9px/1 var(--mono); letter-spacing: .07em; text-transform: uppercase; }
|
|
128
|
+
.stat-value { margin-top: 8px; font-size: 25px; font-weight: 800; letter-spacing: -.035em; }
|
|
129
|
+
.stat-detail { margin-top: 5px; color: var(--muted); font-size: 9px; }
|
|
130
|
+
|
|
131
|
+
.notice {
|
|
132
|
+
display: none;
|
|
133
|
+
margin-top: 14px;
|
|
134
|
+
padding: 13px 16px;
|
|
135
|
+
border: 1px solid var(--line);
|
|
136
|
+
border-left: 3px solid var(--blue);
|
|
137
|
+
border-radius: 7px;
|
|
138
|
+
background: var(--raise);
|
|
139
|
+
color: var(--muted);
|
|
140
|
+
font: 9px/1.65 var(--mono);
|
|
141
|
+
}
|
|
142
|
+
.notice.show { display: block; }
|
|
143
|
+
.notice.warn { border-left-color: var(--amber); color: var(--ink); }
|
|
144
|
+
|
|
145
|
+
.section { margin-top: 64px; }
|
|
146
|
+
.section-head {
|
|
147
|
+
display: flex;
|
|
148
|
+
align-items: center;
|
|
149
|
+
justify-content: space-between;
|
|
150
|
+
gap: 20px;
|
|
151
|
+
margin-bottom: 18px;
|
|
152
|
+
}
|
|
153
|
+
h2 { margin: 0; font-size: 24px; letter-spacing: -.035em; }
|
|
154
|
+
.section-note { margin: 0; color: var(--muted); font-size: 10px; }
|
|
155
|
+
.charts { display: grid; gap: 30px; }
|
|
156
|
+
|
|
157
|
+
.chart-shell {
|
|
158
|
+
position: relative;
|
|
159
|
+
width: 100%;
|
|
160
|
+
min-width: 0;
|
|
161
|
+
padding: 42px 52px 36px;
|
|
162
|
+
overflow: hidden;
|
|
163
|
+
border: 1px solid var(--line);
|
|
164
|
+
border-radius: 22px;
|
|
165
|
+
background: var(--card);
|
|
166
|
+
}
|
|
167
|
+
.chart-header { display: flex; align-items: baseline; gap: 8px; }
|
|
168
|
+
.chart-title { margin: 0; font-size: 24px; font-weight: 800; letter-spacing: -.035em; }
|
|
169
|
+
.chart-subtitle { color: var(--muted); font-size: 22px; font-weight: 700; }
|
|
170
|
+
|
|
171
|
+
.bar-plot {
|
|
172
|
+
display: flex;
|
|
173
|
+
align-items: stretch;
|
|
174
|
+
gap: clamp(18px, 4vw, 54px);
|
|
175
|
+
height: 440px;
|
|
176
|
+
margin-top: 32px;
|
|
177
|
+
padding: 0 24px;
|
|
178
|
+
border-bottom: 2px solid var(--line);
|
|
179
|
+
}
|
|
180
|
+
.bar-item {
|
|
181
|
+
display: grid;
|
|
182
|
+
grid-template-rows: 54px 1fr;
|
|
183
|
+
flex: 1 1 0;
|
|
184
|
+
min-width: 0;
|
|
185
|
+
text-align: center;
|
|
186
|
+
}
|
|
187
|
+
.bar-value {
|
|
188
|
+
align-self: end;
|
|
189
|
+
padding-bottom: 12px;
|
|
190
|
+
color: var(--bar);
|
|
191
|
+
font-size: clamp(22px, 3vw, 38px);
|
|
192
|
+
font-weight: 800;
|
|
193
|
+
line-height: 1;
|
|
194
|
+
letter-spacing: -.05em;
|
|
195
|
+
}
|
|
196
|
+
.bar-item.winner .bar-value { color: var(--green); }
|
|
197
|
+
.bar-slot { display: flex; align-items: end; min-height: 0; }
|
|
198
|
+
.bar {
|
|
199
|
+
width: 100%;
|
|
200
|
+
height: calc(var(--value) * 1%);
|
|
201
|
+
min-height: 3px;
|
|
202
|
+
border-radius: 11px 11px 0 0;
|
|
203
|
+
background: var(--shade, var(--bar));
|
|
204
|
+
transform-origin: bottom;
|
|
205
|
+
animation: rise .55s cubic-bezier(.25,.7,.2,1) both;
|
|
206
|
+
}
|
|
207
|
+
.bar-item.winner .bar { background: linear-gradient(var(--green-soft), var(--green)); }
|
|
208
|
+
@keyframes rise { from { transform: scaleY(0); } }
|
|
209
|
+
.bar-labels {
|
|
210
|
+
display: flex;
|
|
211
|
+
gap: clamp(18px, 4vw, 54px);
|
|
212
|
+
padding: 17px 24px 0;
|
|
213
|
+
}
|
|
214
|
+
.bar-label { flex: 1 1 0; min-width: 0; text-align: center; }
|
|
215
|
+
.model-name { overflow: hidden; font-size: clamp(10px, 1.5vw, 14px); font-weight: 800; text-overflow: ellipsis; white-space: nowrap; }
|
|
216
|
+
.model-meta { margin-top: 5px; color: var(--muted); font-size: clamp(8px, 1vw, 10px); font-weight: 700; }
|
|
217
|
+
|
|
218
|
+
.horizontal-list { display: grid; gap: 20px; margin-top: 46px; }
|
|
219
|
+
.horizontal-row { display: grid; grid-template-columns: 110px 1fr 94px; gap: 18px; align-items: center; }
|
|
220
|
+
.horizontal-name { overflow: hidden; font-size: 12px; font-weight: 800; text-overflow: ellipsis; white-space: nowrap; }
|
|
221
|
+
.horizontal-track { height: 18px; border-radius: 5px; background: var(--track); overflow: hidden; }
|
|
222
|
+
.horizontal-fill { width: calc(var(--value) * 1%); height: 100%; border-radius: inherit; background: var(--bar); }
|
|
223
|
+
.horizontal-row.winner .horizontal-fill { background: var(--green); }
|
|
224
|
+
.horizontal-value { color: var(--bar); font-size: 15px; font-weight: 800; text-align: right; }
|
|
225
|
+
.horizontal-row.winner .horizontal-value { color: var(--green); }
|
|
226
|
+
|
|
227
|
+
.matrix-card { display: flex; flex-direction: column; }
|
|
228
|
+
.matrix-wrap { margin-top: 34px; overflow-x: auto; border: 1px solid var(--line); border-radius: 10px; background: var(--card); }
|
|
229
|
+
table { width: 100%; border-collapse: collapse; font-variant-numeric: tabular-nums; }
|
|
230
|
+
th, td { padding: 13px 14px; border-bottom: 1px solid var(--line); font-size: 10px; text-align: right; white-space: nowrap; }
|
|
231
|
+
tr:last-child td { border-bottom: 0; }
|
|
232
|
+
th { color: var(--muted); background: var(--raise); font: 8px/1 var(--mono); letter-spacing: .07em; text-transform: uppercase; }
|
|
233
|
+
th:first-child, td:first-child { position: sticky; left: 0; text-align: left; }
|
|
234
|
+
th:first-child { z-index: 2; background: var(--raise); }
|
|
235
|
+
td:first-child { z-index: 1; background: var(--card); font-weight: 800; }
|
|
236
|
+
.winner-cell { color: var(--green); font-size: 13px; font-weight: 800; }
|
|
237
|
+
.warning { color: var(--amber); font-weight: 800; }
|
|
238
|
+
.range { display: block; margin-top: 4px; color: var(--muted); font-size: 8px; }
|
|
239
|
+
|
|
240
|
+
.details { display: grid; gap: 8px; }
|
|
241
|
+
details { border: 1px solid var(--line); border-radius: 10px; background: var(--card); overflow: hidden; }
|
|
242
|
+
summary {
|
|
243
|
+
display: grid;
|
|
244
|
+
grid-template-columns: 1fr auto auto;
|
|
245
|
+
gap: 18px;
|
|
246
|
+
align-items: center;
|
|
247
|
+
padding: 17px 19px;
|
|
248
|
+
list-style: none;
|
|
249
|
+
cursor: pointer;
|
|
250
|
+
font-size: 10px;
|
|
251
|
+
}
|
|
252
|
+
summary::-webkit-details-marker { display: none; }
|
|
253
|
+
summary::after { content: "+"; color: var(--muted); font: 15px/1 var(--mono); }
|
|
254
|
+
details[open] summary::after { content: "−"; }
|
|
255
|
+
.detail-title { font-weight: 800; }
|
|
256
|
+
.detail-score { color: var(--muted); }
|
|
257
|
+
.raw-wrap { padding: 0 10px 10px; overflow-x: auto; }
|
|
258
|
+
.raw-wrap table { border-top: 1px solid var(--line); }
|
|
259
|
+
|
|
260
|
+
footer {
|
|
261
|
+
display: flex;
|
|
262
|
+
justify-content: space-between;
|
|
263
|
+
gap: 20px;
|
|
264
|
+
margin-top: 70px;
|
|
265
|
+
padding-top: 24px;
|
|
266
|
+
border-top: 1px solid var(--line);
|
|
267
|
+
color: var(--muted);
|
|
268
|
+
font: 8px/1.4 var(--mono);
|
|
269
|
+
}
|
|
270
|
+
|
|
271
|
+
@media (max-width: 720px) {
|
|
272
|
+
.page { width: min(100% - 24px, 600px); padding-top: 34px; }
|
|
273
|
+
.report-head { display: block; }
|
|
274
|
+
.head-right { align-items: flex-start; margin-top: 24px; }
|
|
275
|
+
.best { text-align: left; }
|
|
276
|
+
.summary { grid-template-columns: repeat(2, 1fr); }
|
|
277
|
+
.stat:nth-child(3) { border-top: 1px solid var(--line); border-left: 0; }
|
|
278
|
+
.stat:nth-child(4) { border-top: 1px solid var(--line); }
|
|
279
|
+
.chart-shell { padding: 25px 18px 24px; border-radius: 14px; }
|
|
280
|
+
.chart-title { font-size: 18px; }
|
|
281
|
+
.chart-subtitle { font-size: 17px; }
|
|
282
|
+
.bar-plot { height: 340px; gap: 12px; padding: 0 4px; }
|
|
283
|
+
.bar-labels { gap: 12px; padding: 14px 4px 0; }
|
|
284
|
+
.horizontal-row { grid-template-columns: 70px 1fr 74px; gap: 10px; }
|
|
285
|
+
}
|
|
286
|
+
</style>
|
|
287
|
+
</head>
|
|
288
|
+
<body>
|
|
289
|
+
<main class="page">
|
|
290
|
+
<header class="report-head">
|
|
291
|
+
<div>
|
|
292
|
+
<h1>BenchKit performance</h1>
|
|
293
|
+
<p class="meta" id="reportMeta"></p>
|
|
294
|
+
</div>
|
|
295
|
+
<div class="head-right">
|
|
296
|
+
<button class="theme-toggle" id="themeToggle" type="button" aria-label="Switch colour theme">
|
|
297
|
+
<span class="theme-dot"></span><span id="themeLabel">Dark</span>
|
|
298
|
+
</button>
|
|
299
|
+
<div class="best" id="best"></div>
|
|
300
|
+
</div>
|
|
301
|
+
</header>
|
|
302
|
+
|
|
303
|
+
<section class="summary" id="summary"></section>
|
|
304
|
+
<div class="notice" id="notice"></div>
|
|
305
|
+
|
|
306
|
+
<section class="section">
|
|
307
|
+
<div class="section-head"><h2>Chart gallery</h2><p class="section-note">Performance across context depth</p></div>
|
|
308
|
+
<div class="charts">
|
|
309
|
+
<div id="generationChart"></div>
|
|
310
|
+
<div id="promptChart"></div>
|
|
311
|
+
<div id="latencyChart"></div>
|
|
312
|
+
</div>
|
|
313
|
+
</section>
|
|
314
|
+
|
|
315
|
+
<section class="section">
|
|
316
|
+
<div class="section-head"><h2>Measurement matrix</h2><p class="section-note">Median with min—max range</p></div>
|
|
317
|
+
<div id="matrix"></div>
|
|
318
|
+
</section>
|
|
319
|
+
|
|
320
|
+
<section class="section">
|
|
321
|
+
<div class="section-head"><h2>Raw repetitions</h2><p class="section-note">Open a depth to inspect every run</p></div>
|
|
322
|
+
<div class="details" id="details"></div>
|
|
323
|
+
</section>
|
|
324
|
+
|
|
325
|
+
<footer><span id="footerMeta"></span><span>Generated by BenchKit</span></footer>
|
|
326
|
+
</main>
|
|
327
|
+
|
|
328
|
+
<script id="perfData" type="application/json">__BENCHKIT_PERF_DATA__</script>
|
|
329
|
+
<script>
|
|
330
|
+
const root = document.documentElement;
|
|
331
|
+
const themeLabel = document.getElementById("themeLabel");
|
|
332
|
+
const setTheme = mode => {
|
|
333
|
+
root.dataset.theme = mode;
|
|
334
|
+
themeLabel.textContent = mode === "dark" ? "Light" : "Dark";
|
|
335
|
+
};
|
|
336
|
+
setTheme(root.dataset.theme === "dark" ? "dark" : "light");
|
|
337
|
+
document.getElementById("themeToggle").addEventListener("click", () => {
|
|
338
|
+
const next = root.dataset.theme === "dark" ? "light" : "dark";
|
|
339
|
+
setTheme(next);
|
|
340
|
+
try { localStorage.setItem("benchkit-theme", next); } catch (error) { /* file:// */ }
|
|
341
|
+
});
|
|
342
|
+
|
|
343
|
+
const profile = JSON.parse(document.getElementById("perfData").textContent);
|
|
344
|
+
const cases = profile.cases || [];
|
|
345
|
+
const settings = profile.settings || {};
|
|
346
|
+
const environment = profile.environment || {};
|
|
347
|
+
const startup = profile.startup || {};
|
|
348
|
+
const drift = profile.drift_check || {};
|
|
349
|
+
const sweep = profile.sweep_summary || {};
|
|
350
|
+
const escapeHtml = value => String(value ?? "").replace(/[&<>"']/g, char => ({"&":"&","<":"<",">":">",'"':""","'":"'"}[char]));
|
|
351
|
+
const number = value => Number(value) || 0;
|
|
352
|
+
const median = (item, key) => number(item?.[key]?.median);
|
|
353
|
+
const metricRange = (item, key, digits = 1) =>
|
|
354
|
+
`${number(item?.[key]?.min).toFixed(digits)}—${number(item?.[key]?.max).toFixed(digits)}`;
|
|
355
|
+
const formatTime = (value, digits = 2) => `${number(value).toFixed(digits)}s`;
|
|
356
|
+
const date = String(profile.generated_at || "").replace("T", " · ");
|
|
357
|
+
const config = environment.config_note || environment.hardware || "";
|
|
358
|
+
const meta = [
|
|
359
|
+
date,
|
|
360
|
+
profile.provider,
|
|
361
|
+
profile.host,
|
|
362
|
+
`${number(settings.gen_tokens)} output tokens`,
|
|
363
|
+
`${number(settings.reps)} measured reps`,
|
|
364
|
+
config,
|
|
365
|
+
].filter(Boolean);
|
|
366
|
+
document.getElementById("reportMeta").textContent = meta.join(" · ");
|
|
367
|
+
document.getElementById("footerMeta").textContent = [profile.model, profile.provider, config].filter(Boolean).join(" · ");
|
|
368
|
+
|
|
369
|
+
const peakTg = Math.max(0, ...cases.map(item => median(item, "tg_tps")));
|
|
370
|
+
document.getElementById("best").innerHTML =
|
|
371
|
+
`<div class="best-score">${peakTg.toFixed(1)}</div><div class="best-model">tok/s · peak TG · ${escapeHtml(profile.model)}</div>`;
|
|
372
|
+
|
|
373
|
+
const summaryItems = [
|
|
374
|
+
["Depths", cases.length, cases.map(item => item.depth_label).join(" · ")],
|
|
375
|
+
["Median TG", `${number(sweep.tg_tps_median).toFixed(1)}`, `tok/s · ${number(sweep.tg_depths)} depths`],
|
|
376
|
+
["Median PP", `${number(sweep.pp_tps_median).toFixed(1)}`, `tok/s · ${number(sweep.pp_depths)} reliable depths`],
|
|
377
|
+
["First request", startup.ttft_s != null ? formatTime(startup.ttft_s, 3) : "—", startup.label || "not captured"],
|
|
378
|
+
];
|
|
379
|
+
document.getElementById("summary").innerHTML = summaryItems.map(item => `
|
|
380
|
+
<article class="stat"><div class="stat-label">${escapeHtml(item[0])}</div>
|
|
381
|
+
<div class="stat-value">${escapeHtml(item[1])}</div><div class="stat-detail">${escapeHtml(item[2])}</div></article>
|
|
382
|
+
`).join("");
|
|
383
|
+
|
|
384
|
+
const calibrated = cases.filter(item => item.depth_method === "calibrated");
|
|
385
|
+
const inaccurate = cases.filter(item => !item.depth_within_tolerance);
|
|
386
|
+
const estimated = cases.filter(item => item.depth_method === "estimated");
|
|
387
|
+
const notices = ["Tip: Run the ruler benchmark to verify how much of the configured context window is actually usable."];
|
|
388
|
+
if (calibrated.length) notices.push("Depth method: calibrated · the server does not expose its native tokenizer route, so server-reported input tokens were used.");
|
|
389
|
+
if (inaccurate.length) notices.push(`⚠ Outside ±1% depth tolerance: ${inaccurate.map(item => item.depth_label).join(", ")}. Use Actual input.`);
|
|
390
|
+
if (estimated.length) notices.push(`⚠ Token-depth calibration unavailable: ${estimated.map(item => item.depth_label).join(", ")}. Use Actual input.`);
|
|
391
|
+
if (drift.error) {
|
|
392
|
+
notices.push(`⚠ End-of-sweep drift check failed: ${drift.error}`);
|
|
393
|
+
} else if (drift.depth_label) {
|
|
394
|
+
notices.push(
|
|
395
|
+
`Drift check · ${drift.depth_label} TG ${number(drift.initial_tg_tps).toFixed(1)} → ${number(drift.final_tg_tps).toFixed(1)} tok/s (${number(drift.change * 100).toFixed(1)}%)`
|
|
396
|
+
+ (drift.stable ? "" : " · possible warmup/drift anomaly")
|
|
397
|
+
);
|
|
398
|
+
}
|
|
399
|
+
const notice = document.getElementById("notice");
|
|
400
|
+
if (notices.length) {
|
|
401
|
+
notice.textContent = notices.join(" ");
|
|
402
|
+
notice.classList.add("show");
|
|
403
|
+
if (inaccurate.length || estimated.length || drift.error || drift.stable === false) notice.classList.add("warn");
|
|
404
|
+
}
|
|
405
|
+
|
|
406
|
+
const shades = [1, 2, 3, 4, 5, 6, 7].map(step => `var(--shade-${step})`);
|
|
407
|
+
const barChart = ({title, subtitle, items, value, meta}) => {
|
|
408
|
+
const maximum = Math.max(1, ...items.map(value));
|
|
409
|
+
const bestValue = Math.max(0, ...items.map(value));
|
|
410
|
+
return `<article class="chart-shell">
|
|
411
|
+
<div class="chart-header"><h3 class="chart-title">${escapeHtml(title)}</h3><span class="chart-subtitle">· ${escapeHtml(subtitle)}</span></div>
|
|
412
|
+
<div class="bar-plot">${items.map((item, index) => {
|
|
413
|
+
const current = value(item);
|
|
414
|
+
const winner = current === bestValue;
|
|
415
|
+
const height = Math.max(0, Math.min(100, current / maximum * 100));
|
|
416
|
+
return `<div class="bar-item ${winner ? "winner" : ""}">
|
|
417
|
+
<div class="bar-value">${current.toFixed(1)}</div>
|
|
418
|
+
<div class="bar-slot"><div class="bar" style="--value:${height};--shade:${shades[Math.min(index, shades.length - 1)]};animation-delay:${index * 45}ms"></div></div>
|
|
419
|
+
</div>`;
|
|
420
|
+
}).join("")}</div>
|
|
421
|
+
<div class="bar-labels">${items.map(item => `<div class="bar-label">
|
|
422
|
+
<div class="model-name">${escapeHtml(item.depth_label)}</div>
|
|
423
|
+
<div class="model-meta">${escapeHtml(meta(item))}</div>
|
|
424
|
+
</div>`).join("")}</div>
|
|
425
|
+
</article>`;
|
|
426
|
+
};
|
|
427
|
+
|
|
428
|
+
document.getElementById("generationChart").innerHTML = barChart({
|
|
429
|
+
title: "Generation speed", subtitle: "TG tok/s", items: cases,
|
|
430
|
+
value: item => median(item, "tg_tps"),
|
|
431
|
+
meta: item => `${number(item.actual_input_tokens).toLocaleString()} input tokens`,
|
|
432
|
+
});
|
|
433
|
+
|
|
434
|
+
const promptCases = cases.filter(item => item.pp_reliable);
|
|
435
|
+
document.getElementById("promptChart").innerHTML = barChart({
|
|
436
|
+
title: "Prompt processing", subtitle: "PP tok/s", items: promptCases,
|
|
437
|
+
value: item => median(item, "pp_tps"),
|
|
438
|
+
meta: item => `${number(item.actual_input_tokens).toLocaleString()} input tokens`,
|
|
439
|
+
});
|
|
440
|
+
|
|
441
|
+
const maxTtft = Math.max(1, ...cases.map(item => median(item, "ttft_s")));
|
|
442
|
+
const fastestTtft = Math.min(...cases.map(item => median(item, "ttft_s")).filter(Boolean));
|
|
443
|
+
document.getElementById("latencyChart").innerHTML = `<article class="chart-shell">
|
|
444
|
+
<div class="chart-header"><h3 class="chart-title">Time to first token</h3><span class="chart-subtitle">· lower is better</span></div>
|
|
445
|
+
<div class="horizontal-list">${cases.map(item => {
|
|
446
|
+
const value = median(item, "ttft_s");
|
|
447
|
+
return `<div class="horizontal-row ${value === fastestTtft ? "winner" : ""}">
|
|
448
|
+
<div class="horizontal-name">${escapeHtml(item.depth_label)}</div>
|
|
449
|
+
<div class="horizontal-track"><div class="horizontal-fill" style="--value:${value / maxTtft * 100}"></div></div>
|
|
450
|
+
<div class="horizontal-value">${formatTime(value, 3)}</div>
|
|
451
|
+
</div>`;
|
|
452
|
+
}).join("")}</div>
|
|
453
|
+
</article>`;
|
|
454
|
+
|
|
455
|
+
const bestTg = Math.max(0, ...cases.map(item => median(item, "tg_tps")));
|
|
456
|
+
document.getElementById("matrix").innerHTML = `<article class="chart-shell matrix-card">
|
|
457
|
+
<div class="chart-header"><h3 class="chart-title">Performance matrix</h3><span class="chart-subtitle">· measured medians</span></div>
|
|
458
|
+
<div class="matrix-wrap"><table>
|
|
459
|
+
<thead><tr><th>Context</th><th>Actual input</th><th>PP tok/s</th><th>TG tok/s</th><th>TTFT</th><th>Wall</th><th>Overhead</th><th>Output</th><th>Runs</th></tr></thead>
|
|
460
|
+
<tbody>${cases.map(item => {
|
|
461
|
+
const warning = item.depth_within_tolerance ? "" : "!";
|
|
462
|
+
const failed = number(item.successful_reps) < number(settings.reps);
|
|
463
|
+
const pp = item.pp_reliable
|
|
464
|
+
? `${median(item, "pp_tps").toFixed(1)}<span class="range">${metricRange(item, "pp_tps")}</span>`
|
|
465
|
+
: `n/a<span class="range">tiny prompt</span>`;
|
|
466
|
+
const overhead = item.overhead_available
|
|
467
|
+
? `${formatTime(median(item, "overhead_s"), 3)}<span class="range">${metricRange(item, "overhead_s", 3)}s</span>`
|
|
468
|
+
: "n/a";
|
|
469
|
+
return `<tr>
|
|
470
|
+
<td class="${warning ? "warning" : ""}">${escapeHtml(item.depth_label)}${warning}</td>
|
|
471
|
+
<td>${number(item.actual_input_tokens).toLocaleString()}</td>
|
|
472
|
+
<td>${pp}</td>
|
|
473
|
+
<td class="${median(item, "tg_tps") === bestTg ? "winner-cell" : ""}">${median(item, "tg_tps").toFixed(1)}<span class="range">${metricRange(item, "tg_tps")}</span></td>
|
|
474
|
+
<td>${formatTime(median(item, "ttft_s"), 3)}<span class="range">${metricRange(item, "ttft_s", 3)}s</span></td>
|
|
475
|
+
<td>${formatTime(median(item, "wall_time_s"), 2)}<span class="range">${metricRange(item, "wall_time_s", 2)}s</span></td>
|
|
476
|
+
<td>${overhead}</td>
|
|
477
|
+
<td>${median(item, "output_tokens").toFixed(0)}</td>
|
|
478
|
+
<td class="${failed ? "warning" : ""}">${number(item.successful_reps)}/${number(settings.reps)}</td>
|
|
479
|
+
</tr>`;
|
|
480
|
+
}).join("")}</tbody>
|
|
481
|
+
</table></div>
|
|
482
|
+
</article>`;
|
|
483
|
+
|
|
484
|
+
document.getElementById("details").innerHTML = cases.map(item => `<details>
|
|
485
|
+
<summary>
|
|
486
|
+
<span class="detail-title">${escapeHtml(item.depth_label)} context · ${number(item.actual_input_tokens).toLocaleString()} actual tokens</span>
|
|
487
|
+
<span class="detail-score">${median(item, "tg_tps").toFixed(1)} TG tok/s · ${formatTime(median(item, "ttft_s"), 3)} TTFT</span>
|
|
488
|
+
</summary>
|
|
489
|
+
<div class="raw-wrap"><table>
|
|
490
|
+
<thead><tr><th>Run</th><th>Input</th><th>Output</th><th>PP tok/s</th><th>TG tok/s</th><th>Wall TG</th><th>TTFT</th><th>Wall</th><th>Overhead</th></tr></thead>
|
|
491
|
+
<tbody>${(item.runs || []).map(run => run.error
|
|
492
|
+
? `<tr><td>${number(run.repetition)}</td><td colspan="8" class="warning">${escapeHtml(run.error)}</td></tr>`
|
|
493
|
+
: `<tr>
|
|
494
|
+
<td>${number(run.repetition)}</td>
|
|
495
|
+
<td>${number(run.input_tokens).toLocaleString()}</td>
|
|
496
|
+
<td>${number(run.output_tokens)}</td>
|
|
497
|
+
<td>${number(run.pp_tps).toFixed(1)}</td>
|
|
498
|
+
<td>${number(run.tg_tps).toFixed(1)}</td>
|
|
499
|
+
<td>${number(run.wall_tg_tps).toFixed(1)}</td>
|
|
500
|
+
<td>${formatTime(run.ttft_s, 3)}</td>
|
|
501
|
+
<td>${formatTime(run.wall_time_s, 2)}</td>
|
|
502
|
+
<td>${run.server_timings ? formatTime(run.overhead_s, 3) : "n/a"}</td>
|
|
503
|
+
</tr>`
|
|
504
|
+
).join("")}</tbody>
|
|
505
|
+
</table></div>
|
|
506
|
+
</details>`).join("");
|
|
507
|
+
</script>
|
|
508
|
+
</body>
|
|
509
|
+
</html>
|