multipleof4/lynchmark
1
0
Fork 0
mirror of https://github.com/multipleof4/lynchmark.git synced 2026-01-13 16:17:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Refactor: Generate browser-runnable test files

Browse Source
This commit is contained in:
multipleof4
2025-10-13 10:24:15 -07:00
parent b47c8499c7
commit fdd4654843
1 changed files with 14 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
scripts/run-benchmark.js
View File

@@ -2,17 +2,12 @@ import { promises as fs } from 'fs';
import path from 'path';
import { pathToFileURL } from 'url';
import axios from 'axios';
import { exec } from 'child_process';
import { promisify } from 'util';
import { performance } from 'perf_hooks';
const execPromise = promisify(exec);
const CWD = process.cwd();
const README_PATH = path.join(CWD, 'README');
const TESTS_DIR = path.join(CWD, 'tests');
const TEMP_FILE = path.join(CWD, 'scripts', 'temp_test.js');
const getLlmCode = async (prompt, model) => {
const getLlmCode = async (prompt, model, functionName) => {
try {
const res = await axios.post(
'https://openrouter.ai/api/v1/chat/completions',
@@ -21,27 +16,13 @@ const getLlmCode = async (prompt, model) => {
);
const content = res.data.choices[0].message.content;
const code = content.match(/```(?:javascript|js)?\n([\s\S]+?)\n```/)?.[1].trim() ?? content.trim();
return code.replace(/^export\s+(default\s+)?/, '');
return `${code.replace(/^export\s+(default\s+)?/, '')}\nexport default ${functionName};`;
} catch (error) {
console.error(`API Error for ${model}: ${error.message}`);
return null;
}
};
const runTest = async (code) => {
const start = performance.now();
let passed = false;
try {
await fs.writeFile(TEMP_FILE, code);
await execPromise(`node ${TEMP_FILE}`);
passed = true;
} catch (error) { // eslint-disable-line no-empty
} finally {
await fs.unlink(TEMP_FILE).catch(() => {});
}
return { passed, duration: (performance.now() - start) / 1000 };
};
const main = async () => {
const readme = await fs.readFile(README_PATH, 'utf-8');
const models = readme.match(/<!-- MODELS_START -->\n([\s\S]+?)\n<!-- MODELS_END -->/)[1].trim().split('\n');
@@ -55,38 +36,21 @@ const main = async () => {
const testsToRun = allTestDirs.slice(0, Math.ceil(allTestDirs.length * (percentage / 100)));
const results = [];
for (const model of models) {
results.push(`**${model}**`);
for (const dir of allTestDirs) {
if (!testsToRun.includes(dir)) {
results.push(`- ${dir}: ⚪ Not Run`);
continue;
}
for (const dir of testsToRun) {
const testModule = await import(pathToFileURL(path.join(TESTS_DIR, dir, 'test.js')));
const { prompt, harness } = testModule.default;
console.log(`Running ${dir} for ${model}...`);
const llmCode = await getLlmCode(prompt, model);
if (llmCode) {
const outDir = path.join(TESTS_DIR, dir, 'outputs');
await fs.mkdir(outDir, { recursive: true });
const fname = `${model.replace(/[\/:]/g, '_')}_${new Date().toISOString().replace(/:/g, '-')}.js`;
await fs.writeFile(path.join(outDir, fname), llmCode);
const { passed, duration } = await runTest(`${llmCode}\n${harness}`);
results.push(`- ${dir}: ${passed ? '✅ Pass' : '❌ Fail'} (${duration.toFixed(3)}s)`);
} else {
results.push(`- ${dir}: ❌ API Error`);
}
}
results.push('');
}
const { prompt, functionName } = testModule.default;
console.log(`Generating ${dir} for ${model}...`);
const llmCode = await getLlmCode(prompt, model, functionName);
if (!llmCode) continue;
const newReadme = readme.replace(
/<!-- RESULTS_START -->[\s\S]*<!-- RESULTS_END -->/,
`<!-- RESULTS_START -->\n${results.join('\n').trim()}\n<!-- RESULTS_END -->`
);
await fs.writeFile(README_PATH, newReadme);
console.log('Benchmark complete. README updated.');
const outDir = path.join(TESTS_DIR, dir, 'outputs');
await fs.mkdir(outDir, { recursive: true });
const fname = `${model.replace(/[\/:]/g, '_')}.js`;
await fs.writeFile(path.join(outDir, fname), llmCode);
}
}
console.log('Code generation complete.');
};
main().catch(console.error);