@darkrishabh/bench-ai 1.0.0
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.
- package/README.md +333 -0
- package/dist/cli/app.d.ts +11 -0
- package/dist/cli/app.d.ts.map +1 -0
- package/dist/cli/app.js +48 -0
- package/dist/cli/app.js.map +1 -0
- package/dist/cli/components/DiffView.d.ts +5 -0
- package/dist/cli/components/DiffView.d.ts.map +1 -0
- package/dist/cli/components/DiffView.js +14 -0
- package/dist/cli/components/DiffView.js.map +1 -0
- package/dist/cli/components/EvalView.d.ts +6 -0
- package/dist/cli/components/EvalView.d.ts.map +1 -0
- package/dist/cli/components/EvalView.js +82 -0
- package/dist/cli/components/EvalView.js.map +1 -0
- package/dist/cli/components/Spinner.d.ts +4 -0
- package/dist/cli/components/Spinner.d.ts.map +1 -0
- package/dist/cli/components/Spinner.js +15 -0
- package/dist/cli/components/Spinner.js.map +1 -0
- package/dist/cli/index.d.ts +3 -0
- package/dist/cli/index.d.ts.map +1 -0
- package/dist/cli/index.js +117 -0
- package/dist/cli/index.js.map +1 -0
- package/dist/cli/run-command.d.ts +11 -0
- package/dist/cli/run-command.d.ts.map +1 -0
- package/dist/cli/run-command.js +119 -0
- package/dist/cli/run-command.js.map +1 -0
- package/dist/engine/cost.d.ts +3 -0
- package/dist/engine/cost.d.ts.map +1 -0
- package/dist/engine/cost.js +52 -0
- package/dist/engine/cost.js.map +1 -0
- package/dist/engine/diff.d.ts +6 -0
- package/dist/engine/diff.d.ts.map +1 -0
- package/dist/engine/diff.js +43 -0
- package/dist/engine/diff.js.map +1 -0
- package/dist/engine/eval.d.ts +14 -0
- package/dist/engine/eval.d.ts.map +1 -0
- package/dist/engine/eval.js +194 -0
- package/dist/engine/eval.js.map +1 -0
- package/dist/engine/index.d.ts +15 -0
- package/dist/engine/index.d.ts.map +1 -0
- package/dist/engine/index.js +10 -0
- package/dist/engine/index.js.map +1 -0
- package/dist/engine/providers/base.d.ts +7 -0
- package/dist/engine/providers/base.d.ts.map +1 -0
- package/dist/engine/providers/base.js +2 -0
- package/dist/engine/providers/base.js.map +1 -0
- package/dist/engine/providers/claude.d.ts +15 -0
- package/dist/engine/providers/claude.d.ts.map +1 -0
- package/dist/engine/providers/claude.js +53 -0
- package/dist/engine/providers/claude.js.map +1 -0
- package/dist/engine/providers/minimax.d.ts +16 -0
- package/dist/engine/providers/minimax.d.ts.map +1 -0
- package/dist/engine/providers/minimax.js +67 -0
- package/dist/engine/providers/minimax.js.map +1 -0
- package/dist/engine/providers/ollama.d.ts +14 -0
- package/dist/engine/providers/ollama.d.ts.map +1 -0
- package/dist/engine/providers/ollama.js +60 -0
- package/dist/engine/providers/ollama.js.map +1 -0
- package/dist/engine/providers/openai-compatible.d.ts +19 -0
- package/dist/engine/providers/openai-compatible.d.ts.map +1 -0
- package/dist/engine/providers/openai-compatible.js +109 -0
- package/dist/engine/providers/openai-compatible.js.map +1 -0
- package/dist/engine/providers/subprocess.d.ts +55 -0
- package/dist/engine/providers/subprocess.d.ts.map +1 -0
- package/dist/engine/providers/subprocess.js +111 -0
- package/dist/engine/providers/subprocess.js.map +1 -0
- package/dist/engine/suite-loader.d.ts +11 -0
- package/dist/engine/suite-loader.d.ts.map +1 -0
- package/dist/engine/suite-loader.js +75 -0
- package/dist/engine/suite-loader.js.map +1 -0
- package/dist/engine/types.d.ts +104 -0
- package/dist/engine/types.d.ts.map +1 -0
- package/dist/engine/types.js +2 -0
- package/dist/engine/types.js.map +1 -0
- package/next-env.d.ts +6 -0
- package/next.config.ts +26 -0
- package/package.json +72 -0
- package/public/icon.svg +14 -0
- package/src/app/api/diff/route.ts +135 -0
- package/src/app/api/models/route.ts +96 -0
- package/src/app/api/suite/route.ts +314 -0
- package/src/app/globals.css +215 -0
- package/src/app/icon.svg +14 -0
- package/src/app/layout.tsx +44 -0
- package/src/app/opengraph-image.tsx +73 -0
- package/src/app/page.tsx +952 -0
- package/src/app/suite/layout.tsx +12 -0
- package/src/app/suite/page.tsx +206 -0
- package/src/app/twitter-image.tsx +1 -0
- package/src/components/BenchAiLogo.tsx +38 -0
- package/src/components/ComparePanel.tsx +643 -0
- package/src/components/ConfigPanel.tsx +809 -0
- package/src/components/MarkdownOutput.tsx +16 -0
- package/src/components/ModelResponseCard.tsx +313 -0
- package/src/components/QuickComparisonBar.tsx +184 -0
- package/src/components/ResponsesLineDiff.tsx +149 -0
- package/src/components/SettingsPanel.tsx +591 -0
- package/src/components/SuitePanel.tsx +875 -0
- package/src/lib/brand.ts +4 -0
- package/src/lib/config-yaml.ts +70 -0
- package/src/lib/consume-suite-sse.ts +70 -0
- package/src/lib/describe-judge.ts +23 -0
- package/src/lib/model-chip-palette.ts +9 -0
- package/src/lib/openai-model-list.ts +33 -0
- package/src/lib/provider-ui.ts +30 -0
- package/src/lib/resolve-credentials.ts +80 -0
- package/src/lib/run-history.ts +66 -0
- package/src/lib/simple-line-diff.ts +50 -0
- package/src/lib/storage.ts +100 -0
- package/src/lib/suite-judge-meta.ts +13 -0
- package/src/lib/suite-run-history.ts +81 -0
- package/src/types.ts +170 -0
- package/vercel.json +5 -0
|
@@ -0,0 +1,643 @@
|
|
|
1
|
+
"use client";
|
|
2
|
+
|
|
3
|
+
import React from "react";
|
|
4
|
+
import type { WebProviderResult } from "../types";
|
|
5
|
+
import { formatCost } from "@darkrishabh/bench-ai";
|
|
6
|
+
|
|
7
|
+
// ─── Helpers ──────────────────────────────────────────────────────────────────
|
|
8
|
+
|
|
9
|
+
function wordCount(text: string) {
|
|
10
|
+
return text.split(/\s+/).filter(Boolean).length;
|
|
11
|
+
}
|
|
12
|
+
|
|
13
|
+
function tokensPerSec(tokens: number, latencyMs: number) {
|
|
14
|
+
return latencyMs > 0 ? Math.round((tokens / latencyMs) * 1000) : 0;
|
|
15
|
+
}
|
|
16
|
+
|
|
17
|
+
const STOP_WORDS = new Set([
|
|
18
|
+
"the","and","for","are","but","not","you","all","can","had","her","was",
|
|
19
|
+
"one","our","out","day","get","has","him","his","how","its","may","new",
|
|
20
|
+
"now","old","see","two","way","who","any","use","this","that","with","have",
|
|
21
|
+
"from","they","know","want","been","good","much","some","time","very","when",
|
|
22
|
+
"come","here","just","like","long","make","many","more","only","over","such",
|
|
23
|
+
"take","than","them","well","were","what","your","into","there","their","will",
|
|
24
|
+
"also","each","which","would","could","should","other","after","about","these",
|
|
25
|
+
"those","being","because","through","between","during","before","without",
|
|
26
|
+
]);
|
|
27
|
+
|
|
28
|
+
function tokenise(text: string): string[] {
|
|
29
|
+
return (text.toLowerCase().match(/\b[a-z]{4,}\b/g) ?? []).filter(
|
|
30
|
+
(w) => !STOP_WORDS.has(w)
|
|
31
|
+
);
|
|
32
|
+
}
|
|
33
|
+
|
|
34
|
+
function jaccardSimilarity(a: string, b: string): number {
|
|
35
|
+
const setA = new Set(tokenise(a));
|
|
36
|
+
const setB = new Set(tokenise(b));
|
|
37
|
+
const intersection = [...setA].filter((w) => setB.has(w)).length;
|
|
38
|
+
const union = new Set([...setA, ...setB]).size;
|
|
39
|
+
return union === 0 ? 0 : intersection / union;
|
|
40
|
+
}
|
|
41
|
+
|
|
42
|
+
function uniqueTerms(text: string, others: string[], n = 8): string[] {
|
|
43
|
+
const freq = new Map<string, number>();
|
|
44
|
+
tokenise(text).forEach((w) => freq.set(w, (freq.get(w) ?? 0) + 1));
|
|
45
|
+
const otherWords = new Set(others.flatMap(tokenise));
|
|
46
|
+
return [...freq.entries()]
|
|
47
|
+
.filter(([w]) => !otherWords.has(w))
|
|
48
|
+
.sort((a, b) => b[1] - a[1])
|
|
49
|
+
.slice(0, n)
|
|
50
|
+
.map(([w]) => w);
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
function countStructure(text: string) {
|
|
54
|
+
return {
|
|
55
|
+
headings: (text.match(/^#{1,6} /gm) ?? []).length,
|
|
56
|
+
bullets: (text.match(/^[-*] /gm) ?? []).length,
|
|
57
|
+
codeBlocks: (text.match(/```/g) ?? []).length / 2,
|
|
58
|
+
tables: (text.match(/^\|.+\|/gm) ?? []).length > 0,
|
|
59
|
+
};
|
|
60
|
+
}
|
|
61
|
+
|
|
62
|
+
// ─── Styles ───────────────────────────────────────────────────────────────────
|
|
63
|
+
|
|
64
|
+
const PROVIDER_COLOR: Record<string, string> = {
|
|
65
|
+
claude: "var(--claude)",
|
|
66
|
+
ollama: "var(--ollama)",
|
|
67
|
+
minimax: "var(--minimax)",
|
|
68
|
+
openai: "var(--openai)",
|
|
69
|
+
groq: "var(--groq)",
|
|
70
|
+
openrouter: "var(--openrouter)",
|
|
71
|
+
"nvidia-nim":"var(--nvidia-nim)",
|
|
72
|
+
together: "var(--together)",
|
|
73
|
+
perplexity: "var(--perplexity)",
|
|
74
|
+
custom: "var(--custom)",
|
|
75
|
+
};
|
|
76
|
+
|
|
77
|
+
const PROVIDER_SUBTLE: Record<string, string> = {
|
|
78
|
+
claude: "var(--claude-subtle)",
|
|
79
|
+
ollama: "var(--ollama-subtle)",
|
|
80
|
+
minimax: "var(--minimax-subtle)",
|
|
81
|
+
openai: "var(--openai-subtle)",
|
|
82
|
+
groq: "var(--groq-subtle)",
|
|
83
|
+
openrouter: "var(--openrouter-subtle)",
|
|
84
|
+
"nvidia-nim":"var(--nvidia-nim-subtle)",
|
|
85
|
+
together: "var(--together-subtle)",
|
|
86
|
+
perplexity: "var(--perplexity-subtle)",
|
|
87
|
+
custom: "var(--custom-subtle)",
|
|
88
|
+
};
|
|
89
|
+
|
|
90
|
+
// ─── Sub-components ───────────────────────────────────────────────────────────
|
|
91
|
+
|
|
92
|
+
function Section({
|
|
93
|
+
title,
|
|
94
|
+
children,
|
|
95
|
+
}: {
|
|
96
|
+
title: string;
|
|
97
|
+
children: React.ReactNode;
|
|
98
|
+
}) {
|
|
99
|
+
return (
|
|
100
|
+
<div
|
|
101
|
+
style={{
|
|
102
|
+
background: "var(--surface)",
|
|
103
|
+
border: "1px solid var(--border)",
|
|
104
|
+
borderRadius: "var(--r-xl)",
|
|
105
|
+
overflow: "hidden",
|
|
106
|
+
boxShadow: "var(--shadow-sm)",
|
|
107
|
+
}}
|
|
108
|
+
>
|
|
109
|
+
<div
|
|
110
|
+
style={{
|
|
111
|
+
padding: "0.75rem 1.2rem",
|
|
112
|
+
borderBottom: "1px solid var(--border)",
|
|
113
|
+
fontWeight: 600,
|
|
114
|
+
fontSize: "0.8125rem",
|
|
115
|
+
letterSpacing: "-0.01em",
|
|
116
|
+
color: "var(--text-1)",
|
|
117
|
+
background: "var(--surface-subtle)",
|
|
118
|
+
}}
|
|
119
|
+
>
|
|
120
|
+
{title}
|
|
121
|
+
</div>
|
|
122
|
+
<div style={{ padding: "1.25rem" }}>{children}</div>
|
|
123
|
+
</div>
|
|
124
|
+
);
|
|
125
|
+
}
|
|
126
|
+
|
|
127
|
+
function WinnerCard({
|
|
128
|
+
icon,
|
|
129
|
+
label,
|
|
130
|
+
winner,
|
|
131
|
+
value,
|
|
132
|
+
color,
|
|
133
|
+
}: {
|
|
134
|
+
icon: string;
|
|
135
|
+
label: string;
|
|
136
|
+
winner: string;
|
|
137
|
+
value: string;
|
|
138
|
+
color: string;
|
|
139
|
+
}) {
|
|
140
|
+
return (
|
|
141
|
+
<div
|
|
142
|
+
style={{
|
|
143
|
+
flex: 1,
|
|
144
|
+
background: "var(--surface)",
|
|
145
|
+
border: "1px solid var(--border)",
|
|
146
|
+
borderTop: `3px solid ${color}`,
|
|
147
|
+
borderRadius: "var(--r-md)",
|
|
148
|
+
padding: "1rem 1.1rem",
|
|
149
|
+
boxShadow: "var(--shadow-xs)",
|
|
150
|
+
minWidth: 0,
|
|
151
|
+
}}
|
|
152
|
+
>
|
|
153
|
+
<div
|
|
154
|
+
style={{
|
|
155
|
+
fontSize: "0.72rem",
|
|
156
|
+
textTransform: "uppercase",
|
|
157
|
+
letterSpacing: "0.06em",
|
|
158
|
+
color: "var(--text-3)",
|
|
159
|
+
marginBottom: "0.35rem",
|
|
160
|
+
}}
|
|
161
|
+
>
|
|
162
|
+
{icon} {label}
|
|
163
|
+
</div>
|
|
164
|
+
<div
|
|
165
|
+
style={{
|
|
166
|
+
fontWeight: 700,
|
|
167
|
+
fontSize: "0.95rem",
|
|
168
|
+
color: "var(--text-1)",
|
|
169
|
+
whiteSpace: "nowrap",
|
|
170
|
+
overflow: "hidden",
|
|
171
|
+
textOverflow: "ellipsis",
|
|
172
|
+
marginBottom: "0.2rem",
|
|
173
|
+
}}
|
|
174
|
+
>
|
|
175
|
+
{winner}
|
|
176
|
+
</div>
|
|
177
|
+
<div style={{ fontSize: "0.8rem", color, fontWeight: 600 }}>{value}</div>
|
|
178
|
+
</div>
|
|
179
|
+
);
|
|
180
|
+
}
|
|
181
|
+
|
|
182
|
+
function MetricRow({
|
|
183
|
+
label,
|
|
184
|
+
results,
|
|
185
|
+
getValue,
|
|
186
|
+
format,
|
|
187
|
+
lowerIsBetter,
|
|
188
|
+
}: {
|
|
189
|
+
label: string;
|
|
190
|
+
results: WebProviderResult[];
|
|
191
|
+
getValue: (r: WebProviderResult) => number;
|
|
192
|
+
format: (v: number) => string;
|
|
193
|
+
lowerIsBetter?: boolean;
|
|
194
|
+
}) {
|
|
195
|
+
const values = results.map(getValue);
|
|
196
|
+
const max = Math.max(...values);
|
|
197
|
+
const min = Math.min(...values);
|
|
198
|
+
const winnerIdx = lowerIsBetter
|
|
199
|
+
? values.indexOf(min)
|
|
200
|
+
: values.indexOf(max);
|
|
201
|
+
|
|
202
|
+
return (
|
|
203
|
+
<div style={{ display: "flex", flexDirection: "column", gap: "0.5rem" }}>
|
|
204
|
+
<div
|
|
205
|
+
style={{
|
|
206
|
+
fontSize: "0.75rem",
|
|
207
|
+
fontWeight: 600,
|
|
208
|
+
color: "var(--text-2)",
|
|
209
|
+
textTransform: "uppercase",
|
|
210
|
+
letterSpacing: "0.04em",
|
|
211
|
+
}}
|
|
212
|
+
>
|
|
213
|
+
{label}
|
|
214
|
+
</div>
|
|
215
|
+
{results.map((r, i) => {
|
|
216
|
+
const v = getValue(r);
|
|
217
|
+
const pct = max > 0 ? (v / max) * 100 : 0;
|
|
218
|
+
const barPct = lowerIsBetter && max > 0 ? (v / max) * 100 : pct;
|
|
219
|
+
const barColor =
|
|
220
|
+
i === winnerIdx
|
|
221
|
+
? PROVIDER_COLOR[r.provider] ?? "var(--accent)"
|
|
222
|
+
: "var(--border-strong)";
|
|
223
|
+
const isWinner = i === winnerIdx;
|
|
224
|
+
|
|
225
|
+
return (
|
|
226
|
+
<div key={r.instanceId} style={{ display: "flex", alignItems: "center", gap: "0.75rem" }}>
|
|
227
|
+
<div
|
|
228
|
+
style={{
|
|
229
|
+
width: 130,
|
|
230
|
+
fontSize: "0.78rem",
|
|
231
|
+
color: "var(--text-2)",
|
|
232
|
+
flexShrink: 0,
|
|
233
|
+
overflow: "hidden",
|
|
234
|
+
textOverflow: "ellipsis",
|
|
235
|
+
whiteSpace: "nowrap",
|
|
236
|
+
}}
|
|
237
|
+
title={r.label}
|
|
238
|
+
>
|
|
239
|
+
{r.label}
|
|
240
|
+
</div>
|
|
241
|
+
<div
|
|
242
|
+
style={{
|
|
243
|
+
flex: 1,
|
|
244
|
+
height: 8,
|
|
245
|
+
background: "var(--surface-subtle)",
|
|
246
|
+
borderRadius: 4,
|
|
247
|
+
overflow: "hidden",
|
|
248
|
+
border: "1px solid var(--border)",
|
|
249
|
+
}}
|
|
250
|
+
>
|
|
251
|
+
<div
|
|
252
|
+
style={{
|
|
253
|
+
width: `${barPct}%`,
|
|
254
|
+
height: "100%",
|
|
255
|
+
background: barColor,
|
|
256
|
+
borderRadius: 4,
|
|
257
|
+
transition: "width 0.4s ease",
|
|
258
|
+
}}
|
|
259
|
+
/>
|
|
260
|
+
</div>
|
|
261
|
+
<div
|
|
262
|
+
style={{
|
|
263
|
+
fontSize: "0.8rem",
|
|
264
|
+
fontWeight: isWinner ? 600 : 400,
|
|
265
|
+
color: isWinner
|
|
266
|
+
? PROVIDER_COLOR[r.provider] ?? "var(--accent)"
|
|
267
|
+
: "var(--text-2)",
|
|
268
|
+
width: 90,
|
|
269
|
+
textAlign: "right",
|
|
270
|
+
flexShrink: 0,
|
|
271
|
+
display: "flex",
|
|
272
|
+
alignItems: "center",
|
|
273
|
+
justifyContent: "flex-end",
|
|
274
|
+
gap: "0.3rem",
|
|
275
|
+
}}
|
|
276
|
+
>
|
|
277
|
+
{format(v)}
|
|
278
|
+
{isWinner && (
|
|
279
|
+
<span
|
|
280
|
+
style={{
|
|
281
|
+
fontSize: "0.65rem",
|
|
282
|
+
background: PROVIDER_SUBTLE[r.provider] ?? "var(--accent-subtle)",
|
|
283
|
+
color: PROVIDER_COLOR[r.provider] ?? "var(--accent)",
|
|
284
|
+
border: `1px solid ${PROVIDER_COLOR[r.provider] ?? "var(--accent)"}40`,
|
|
285
|
+
borderRadius: 4,
|
|
286
|
+
padding: "0.05rem 0.35rem",
|
|
287
|
+
fontWeight: 700,
|
|
288
|
+
textTransform: "uppercase",
|
|
289
|
+
letterSpacing: "0.03em",
|
|
290
|
+
}}
|
|
291
|
+
>
|
|
292
|
+
{lowerIsBetter ? "best" : "most"}
|
|
293
|
+
</span>
|
|
294
|
+
)}
|
|
295
|
+
</div>
|
|
296
|
+
</div>
|
|
297
|
+
);
|
|
298
|
+
})}
|
|
299
|
+
</div>
|
|
300
|
+
);
|
|
301
|
+
}
|
|
302
|
+
|
|
303
|
+
function SimilarityPill({ pct }: { pct: number }) {
|
|
304
|
+
const color =
|
|
305
|
+
pct > 0.65 ? "var(--green)" : pct > 0.4 ? "var(--amber)" : "var(--text-3)";
|
|
306
|
+
return (
|
|
307
|
+
<div
|
|
308
|
+
style={{
|
|
309
|
+
display: "inline-flex",
|
|
310
|
+
alignItems: "center",
|
|
311
|
+
gap: "0.5rem",
|
|
312
|
+
background: "var(--surface-subtle)",
|
|
313
|
+
border: "1px solid var(--border)",
|
|
314
|
+
borderRadius: 8,
|
|
315
|
+
padding: "0.5rem 0.75rem",
|
|
316
|
+
width: "100%",
|
|
317
|
+
}}
|
|
318
|
+
>
|
|
319
|
+
<div style={{ flex: 1, height: 6, background: "var(--border)", borderRadius: 3, overflow: "hidden" }}>
|
|
320
|
+
<div style={{ width: `${pct * 100}%`, height: "100%", background: color, borderRadius: 3, transition: "width 0.4s ease" }} />
|
|
321
|
+
</div>
|
|
322
|
+
<span style={{ fontSize: "0.82rem", fontWeight: 700, color, width: 36, textAlign: "right", flexShrink: 0 }}>
|
|
323
|
+
{Math.round(pct * 100)}%
|
|
324
|
+
</span>
|
|
325
|
+
</div>
|
|
326
|
+
);
|
|
327
|
+
}
|
|
328
|
+
|
|
329
|
+
function TermChip({ term, color }: { term: string; color: string }) {
|
|
330
|
+
return (
|
|
331
|
+
<span
|
|
332
|
+
style={{
|
|
333
|
+
padding: "0.2rem 0.55rem",
|
|
334
|
+
borderRadius: 5,
|
|
335
|
+
fontSize: "0.75rem",
|
|
336
|
+
background: color + "12",
|
|
337
|
+
border: `1px solid ${color}28`,
|
|
338
|
+
color,
|
|
339
|
+
fontWeight: 500,
|
|
340
|
+
}}
|
|
341
|
+
>
|
|
342
|
+
{term}
|
|
343
|
+
</span>
|
|
344
|
+
);
|
|
345
|
+
}
|
|
346
|
+
|
|
347
|
+
// ─── Main component ───────────────────────────────────────────────────────────
|
|
348
|
+
|
|
349
|
+
export function ComparePanel({ results }: { results: WebProviderResult[] }) {
|
|
350
|
+
const valid = results.filter((r) => !r.error && r.output.length > 0);
|
|
351
|
+
|
|
352
|
+
if (valid.length < 2) {
|
|
353
|
+
return (
|
|
354
|
+
<div
|
|
355
|
+
style={{
|
|
356
|
+
textAlign: "center",
|
|
357
|
+
padding: "3rem",
|
|
358
|
+
color: "var(--text-3)",
|
|
359
|
+
fontSize: "0.875rem",
|
|
360
|
+
}}
|
|
361
|
+
>
|
|
362
|
+
Need at least 2 successful responses to compare.
|
|
363
|
+
</div>
|
|
364
|
+
);
|
|
365
|
+
}
|
|
366
|
+
|
|
367
|
+
// ── At a glance ──
|
|
368
|
+
const fastest = valid.reduce((a, b) => (a.latencyMs < b.latencyMs ? a : b));
|
|
369
|
+
const cheapest = valid.reduce((a, b) => (a.costUsd <= b.costUsd ? a : b));
|
|
370
|
+
const mostDetailed = valid.reduce((a, b) =>
|
|
371
|
+
wordCount(a.output) >= wordCount(b.output) ? a : b
|
|
372
|
+
);
|
|
373
|
+
const tokensPerSecValues = valid.map((r) => tokensPerSec(r.outputTokens, r.latencyMs));
|
|
374
|
+
const fastestGenIdx = tokensPerSecValues.indexOf(Math.max(...tokensPerSecValues));
|
|
375
|
+
const fastestGen = valid[fastestGenIdx];
|
|
376
|
+
|
|
377
|
+
// ── Similarity pairs ──
|
|
378
|
+
const pairs: { a: WebProviderResult; b: WebProviderResult; sim: number }[] = [];
|
|
379
|
+
for (let i = 0; i < valid.length; i++) {
|
|
380
|
+
for (let j = i + 1; j < valid.length; j++) {
|
|
381
|
+
pairs.push({
|
|
382
|
+
a: valid[i],
|
|
383
|
+
b: valid[j],
|
|
384
|
+
sim: jaccardSimilarity(valid[i].output, valid[j].output),
|
|
385
|
+
});
|
|
386
|
+
}
|
|
387
|
+
}
|
|
388
|
+
|
|
389
|
+
// ── Unique terms ──
|
|
390
|
+
const uniqueByModel = valid.map((r) =>
|
|
391
|
+
uniqueTerms(
|
|
392
|
+
r.output,
|
|
393
|
+
valid.filter((o) => o.instanceId !== r.instanceId).map((o) => o.output)
|
|
394
|
+
)
|
|
395
|
+
);
|
|
396
|
+
|
|
397
|
+
// ── Structure ──
|
|
398
|
+
const structures = valid.map((r) => countStructure(r.output));
|
|
399
|
+
|
|
400
|
+
return (
|
|
401
|
+
<div style={{ display: "flex", flexDirection: "column", gap: "1rem" }}>
|
|
402
|
+
{/* At a glance */}
|
|
403
|
+
<div style={{ display: "flex", gap: "0.75rem", flexWrap: "wrap" }}>
|
|
404
|
+
<WinnerCard
|
|
405
|
+
icon="⚡"
|
|
406
|
+
label="Fastest"
|
|
407
|
+
winner={fastest.label}
|
|
408
|
+
value={`${fastest.latencyMs.toLocaleString()}ms`}
|
|
409
|
+
color={PROVIDER_COLOR[fastest.provider] ?? "var(--accent)"}
|
|
410
|
+
/>
|
|
411
|
+
<WinnerCard
|
|
412
|
+
icon="💰"
|
|
413
|
+
label="Cheapest"
|
|
414
|
+
winner={cheapest.label}
|
|
415
|
+
value={cheapest.costUsd === 0 ? "Free (local)" : formatCost(cheapest.costUsd)}
|
|
416
|
+
color={PROVIDER_COLOR[cheapest.provider] ?? "var(--accent)"}
|
|
417
|
+
/>
|
|
418
|
+
<WinnerCard
|
|
419
|
+
icon="📝"
|
|
420
|
+
label="Most Detailed"
|
|
421
|
+
winner={mostDetailed.label}
|
|
422
|
+
value={`${wordCount(mostDetailed.output).toLocaleString()} words`}
|
|
423
|
+
color={PROVIDER_COLOR[mostDetailed.provider] ?? "var(--accent)"}
|
|
424
|
+
/>
|
|
425
|
+
{fastestGen && tokensPerSecValues[fastestGenIdx] > 0 && (
|
|
426
|
+
<WinnerCard
|
|
427
|
+
icon="🔥"
|
|
428
|
+
label="Fastest Generation"
|
|
429
|
+
winner={fastestGen.label}
|
|
430
|
+
value={`${tokensPerSecValues[fastestGenIdx]} tok/s`}
|
|
431
|
+
color={PROVIDER_COLOR[fastestGen.provider] ?? "var(--accent)"}
|
|
432
|
+
/>
|
|
433
|
+
)}
|
|
434
|
+
</div>
|
|
435
|
+
|
|
436
|
+
{/* Performance metrics */}
|
|
437
|
+
<Section title="Performance">
|
|
438
|
+
<div style={{ display: "flex", flexDirection: "column", gap: "1.25rem" }}>
|
|
439
|
+
<MetricRow
|
|
440
|
+
label="Latency"
|
|
441
|
+
results={valid}
|
|
442
|
+
getValue={(r) => r.latencyMs}
|
|
443
|
+
format={(v) => `${v.toLocaleString()}ms`}
|
|
444
|
+
lowerIsBetter
|
|
445
|
+
/>
|
|
446
|
+
<MetricRow
|
|
447
|
+
label="Cost"
|
|
448
|
+
results={valid}
|
|
449
|
+
getValue={(r) => r.costUsd}
|
|
450
|
+
format={(v) => (v === 0 ? "$0.00" : formatCost(v))}
|
|
451
|
+
lowerIsBetter
|
|
452
|
+
/>
|
|
453
|
+
{valid.some((r) => r.outputTokens > 0) && (
|
|
454
|
+
<MetricRow
|
|
455
|
+
label="Generation Speed (tok/s)"
|
|
456
|
+
results={valid}
|
|
457
|
+
getValue={(r) => tokensPerSec(r.outputTokens, r.latencyMs)}
|
|
458
|
+
format={(v) => (v > 0 ? `${v} tok/s` : "—")}
|
|
459
|
+
/>
|
|
460
|
+
)}
|
|
461
|
+
</div>
|
|
462
|
+
</Section>
|
|
463
|
+
|
|
464
|
+
{/* Output analysis */}
|
|
465
|
+
<Section title="Output Analysis">
|
|
466
|
+
<div style={{ display: "flex", flexDirection: "column", gap: "1.25rem" }}>
|
|
467
|
+
<MetricRow
|
|
468
|
+
label="Word Count"
|
|
469
|
+
results={valid}
|
|
470
|
+
getValue={(r) => wordCount(r.output)}
|
|
471
|
+
format={(v) => `${v.toLocaleString()} words`}
|
|
472
|
+
/>
|
|
473
|
+
{valid.some((r) => r.outputTokens > 0) && (
|
|
474
|
+
<MetricRow
|
|
475
|
+
label="Output Tokens"
|
|
476
|
+
results={valid}
|
|
477
|
+
getValue={(r) => r.outputTokens}
|
|
478
|
+
format={(v) => (v > 0 ? `${v.toLocaleString()} tokens` : "—")}
|
|
479
|
+
/>
|
|
480
|
+
)}
|
|
481
|
+
</div>
|
|
482
|
+
|
|
483
|
+
{/* Structure signals */}
|
|
484
|
+
<div
|
|
485
|
+
style={{
|
|
486
|
+
marginTop: "1.25rem",
|
|
487
|
+
paddingTop: "1.25rem",
|
|
488
|
+
borderTop: "1px solid var(--border)",
|
|
489
|
+
display: "flex",
|
|
490
|
+
flexDirection: "column",
|
|
491
|
+
gap: "0.6rem",
|
|
492
|
+
}}
|
|
493
|
+
>
|
|
494
|
+
<div
|
|
495
|
+
style={{
|
|
496
|
+
fontSize: "0.75rem",
|
|
497
|
+
fontWeight: 600,
|
|
498
|
+
color: "var(--text-2)",
|
|
499
|
+
textTransform: "uppercase",
|
|
500
|
+
letterSpacing: "0.04em",
|
|
501
|
+
marginBottom: "0.25rem",
|
|
502
|
+
}}
|
|
503
|
+
>
|
|
504
|
+
Structure
|
|
505
|
+
</div>
|
|
506
|
+
{valid.map((r, i) => {
|
|
507
|
+
const s = structures[i];
|
|
508
|
+
const tags = [
|
|
509
|
+
s.headings > 0 && `${s.headings} heading${s.headings > 1 ? "s" : ""}`,
|
|
510
|
+
s.bullets > 0 && `${s.bullets} bullet${s.bullets > 1 ? "s" : ""}`,
|
|
511
|
+
s.codeBlocks > 0 && `${Math.round(s.codeBlocks)} code block${s.codeBlocks > 1 ? "s" : ""}`,
|
|
512
|
+
s.tables && "table",
|
|
513
|
+
].filter(Boolean);
|
|
514
|
+
|
|
515
|
+
return (
|
|
516
|
+
<div
|
|
517
|
+
key={r.instanceId}
|
|
518
|
+
style={{ display: "flex", alignItems: "center", gap: "0.6rem" }}
|
|
519
|
+
>
|
|
520
|
+
<div
|
|
521
|
+
style={{
|
|
522
|
+
width: 130,
|
|
523
|
+
fontSize: "0.78rem",
|
|
524
|
+
color: "var(--text-2)",
|
|
525
|
+
flexShrink: 0,
|
|
526
|
+
overflow: "hidden",
|
|
527
|
+
textOverflow: "ellipsis",
|
|
528
|
+
whiteSpace: "nowrap",
|
|
529
|
+
}}
|
|
530
|
+
>
|
|
531
|
+
{r.label}
|
|
532
|
+
</div>
|
|
533
|
+
<div style={{ display: "flex", gap: "0.35rem", flexWrap: "wrap" }}>
|
|
534
|
+
{tags.length > 0 ? (
|
|
535
|
+
tags.map((tag) => (
|
|
536
|
+
<span
|
|
537
|
+
key={String(tag)}
|
|
538
|
+
style={{
|
|
539
|
+
padding: "0.1rem 0.5rem",
|
|
540
|
+
borderRadius: 5,
|
|
541
|
+
fontSize: "0.75rem",
|
|
542
|
+
background: "var(--surface-subtle)",
|
|
543
|
+
border: "1px solid var(--border)",
|
|
544
|
+
color: "var(--text-2)",
|
|
545
|
+
}}
|
|
546
|
+
>
|
|
547
|
+
{tag}
|
|
548
|
+
</span>
|
|
549
|
+
))
|
|
550
|
+
) : (
|
|
551
|
+
<span style={{ fontSize: "0.78rem", color: "var(--text-3)" }}>
|
|
552
|
+
plain prose
|
|
553
|
+
</span>
|
|
554
|
+
)}
|
|
555
|
+
</div>
|
|
556
|
+
</div>
|
|
557
|
+
);
|
|
558
|
+
})}
|
|
559
|
+
</div>
|
|
560
|
+
</Section>
|
|
561
|
+
|
|
562
|
+
{/* Similarity */}
|
|
563
|
+
{pairs.length > 0 && (
|
|
564
|
+
<Section title="Similarity">
|
|
565
|
+
<div style={{ display: "flex", flexDirection: "column", gap: "0.75rem" }}>
|
|
566
|
+
{pairs.map((p) => (
|
|
567
|
+
<div
|
|
568
|
+
key={`${p.a.instanceId}-${p.b.instanceId}`}
|
|
569
|
+
style={{ display: "flex", alignItems: "center", gap: "0.75rem" }}
|
|
570
|
+
>
|
|
571
|
+
<div
|
|
572
|
+
style={{
|
|
573
|
+
fontSize: "0.78rem",
|
|
574
|
+
color: "var(--text-2)",
|
|
575
|
+
flexShrink: 0,
|
|
576
|
+
width: 260,
|
|
577
|
+
overflow: "hidden",
|
|
578
|
+
textOverflow: "ellipsis",
|
|
579
|
+
whiteSpace: "nowrap",
|
|
580
|
+
}}
|
|
581
|
+
>
|
|
582
|
+
<span style={{ color: PROVIDER_COLOR[p.a.provider] ?? "inherit", fontWeight: 500 }}>
|
|
583
|
+
{p.a.label}
|
|
584
|
+
</span>
|
|
585
|
+
<span style={{ color: "var(--text-3)", margin: "0 0.35rem" }}>↔</span>
|
|
586
|
+
<span style={{ color: PROVIDER_COLOR[p.b.provider] ?? "inherit", fontWeight: 500 }}>
|
|
587
|
+
{p.b.label}
|
|
588
|
+
</span>
|
|
589
|
+
</div>
|
|
590
|
+
<SimilarityPill pct={p.sim} />
|
|
591
|
+
</div>
|
|
592
|
+
))}
|
|
593
|
+
<p style={{ fontSize: "0.75rem", color: "var(--text-3)", marginTop: "0.25rem" }}>
|
|
594
|
+
Jaccard similarity on meaningful word overlap. >65% = largely similar, <40% = notably different.
|
|
595
|
+
</p>
|
|
596
|
+
</div>
|
|
597
|
+
</Section>
|
|
598
|
+
)}
|
|
599
|
+
|
|
600
|
+
{/* Unique contributions */}
|
|
601
|
+
{uniqueByModel.some((terms) => terms.length > 0) && (
|
|
602
|
+
<Section title="Unique Contributions">
|
|
603
|
+
<div style={{ display: "flex", flexDirection: "column", gap: "1rem" }}>
|
|
604
|
+
{valid.map((r, i) => {
|
|
605
|
+
const terms = uniqueByModel[i];
|
|
606
|
+
const color = PROVIDER_COLOR[r.provider] ?? "var(--accent)";
|
|
607
|
+
return (
|
|
608
|
+
<div key={r.instanceId} style={{ display: "flex", gap: "0.75rem", alignItems: "flex-start" }}>
|
|
609
|
+
<div
|
|
610
|
+
style={{
|
|
611
|
+
fontSize: "0.78rem",
|
|
612
|
+
color: "var(--text-2)",
|
|
613
|
+
width: 130,
|
|
614
|
+
flexShrink: 0,
|
|
615
|
+
paddingTop: "0.15rem",
|
|
616
|
+
overflow: "hidden",
|
|
617
|
+
textOverflow: "ellipsis",
|
|
618
|
+
whiteSpace: "nowrap",
|
|
619
|
+
}}
|
|
620
|
+
>
|
|
621
|
+
{r.label}
|
|
622
|
+
</div>
|
|
623
|
+
<div style={{ display: "flex", gap: "0.35rem", flexWrap: "wrap", flex: 1 }}>
|
|
624
|
+
{terms.length > 0 ? (
|
|
625
|
+
terms.map((t) => <TermChip key={t} term={t} color={color} />)
|
|
626
|
+
) : (
|
|
627
|
+
<span style={{ fontSize: "0.78rem", color: "var(--text-3)" }}>
|
|
628
|
+
No notably unique terms
|
|
629
|
+
</span>
|
|
630
|
+
)}
|
|
631
|
+
</div>
|
|
632
|
+
</div>
|
|
633
|
+
);
|
|
634
|
+
})}
|
|
635
|
+
<p style={{ fontSize: "0.75rem", color: "var(--text-3)", marginTop: "0.25rem" }}>
|
|
636
|
+
Terms that appear in this model's response but not in others — indicating unique focus areas.
|
|
637
|
+
</p>
|
|
638
|
+
</div>
|
|
639
|
+
</Section>
|
|
640
|
+
)}
|
|
641
|
+
</div>
|
|
642
|
+
);
|
|
643
|
+
}
|