github/toon
1
0
Fork 0
mirror of https://github.com/voson-wang/toon.git synced 2026-01-29 23:34:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

chore: split token efficiency benchmark into mixed/flat tracks

Browse Source
This commit is contained in:
Johann Schopplich
2025-11-06 22:17:18 +01:00
parent e22884308b
commit 54433de930
13 changed files with 567 additions and 1830 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
benchmarks/src/report.ts
View File

@@ -80,8 +80,13 @@ export function generateAccuracyReport(
return `
Benchmarks test LLM comprehension across different input formats using ${totalQuestions} data retrieval questions on ${modelNames.length} ${modelNames.length === 1 ? 'model' : 'models'}.
<details>
<summary><strong>View Dataset Catalog</strong></summary>
${generateDatasetCatalog(ACCURACY_DATASETS)}
</details>
#### Efficiency Ranking (Accuracy per 1K Tokens)
${generateEfficiencyRankingReport(formatResults)}
@@ -118,7 +123,7 @@ ${rows}
- **nested**: Objects with nested structures (nested objects or arrays)
- **deep**: Highly nested with minimal tabular eligibility
**CSV Support:** ✓ (supported), ✗ (not supported - would require lossy flattening)
**CSV Support:** ✓ (supported), ✗ (not supported – would require lossy flattening)
**Eligibility:** Percentage of arrays that qualify for TOON's tabular format (uniform objects with primitive values)
`.trim()
@@ -219,7 +224,7 @@ function generateDetailedAccuracyReport(
const totalEvaluations = totalQuestions * formatCount * modelNames.length
return `
Accuracy across **${modelNames.length} ${modelNames.length === 1 ? 'LLM' : 'LLMs'}** on ${totalQuestions} data retrieval questions:
Accuracy across ${modelNames.length} ${modelNames.length === 1 ? 'LLM' : 'LLMs'} on ${totalQuestions} data retrieval questions:
\`\`\`
${modelBreakdown}
@@ -453,13 +458,17 @@ function generateHorizontalEfficiencyChart(
): string {
const barWidth = 20
const maxEfficiency = Math.max(...ranking.map(r => r.efficiency))
const maxFormatWidth = Math.max(...ranking.map(r => r.format.length))
const maxFormatWidth = Math.max(...ranking.map((r) => {
const displayName = FORMATTER_DISPLAY_NAMES[r.format] || r.format
return displayName.length
}))
return ranking
.map((r) => {
const normalizedValue = r.efficiency / maxEfficiency
const bar = createProgressBar(normalizedValue, 1, barWidth, { filled: '▓', empty: '░' })
const formatName = r.format.padEnd(maxFormatWidth)
const displayName = FORMATTER_DISPLAY_NAMES[r.format] || r.format
const formatName = displayName.padEnd(maxFormatWidth)
const efficiency = r.efficiency.toFixed(1).padStart(4)
const accuracy = `${(r.accuracy * 100).toFixed(1)}%`.padStart(5)
const tokens = r.tokens.toLocaleString('en-US').padStart(5)