browser-speak / tools /run-webgpu-benchmark.mjs
Mike0021's picture
Add worker network telemetry to browser evidence
d2ae80e verified
#!/usr/bin/env node
import { spawn } from "node:child_process";
import { existsSync } from "node:fs";
import { mkdir, rm, writeFile } from "node:fs/promises";
import { tmpdir } from "node:os";
import { dirname, resolve } from "node:path";
import { sourceFingerprint } from "./source-fingerprint.mjs";
const url = process.env.BROWSER_SPEAK_URL ?? "http://127.0.0.1:5174/";
const chrome =
process.env.CHROME_BIN ??
(existsSync("/opt/google/chrome/chrome")
? "/opt/google/chrome/chrome"
: existsSync("/usr/bin/google-chrome")
? "/usr/bin/google-chrome"
: "chromium");
const resultPath = resolve(process.env.BROWSER_SPEAK_WEBGPU_JSON ?? `${tmpdir()}/browser-speak-webgpu-results.json`);
const profileDir = resolve(process.env.BROWSER_SPEAK_WEBGPU_PROFILE_DIR ?? `${tmpdir()}/browser-speak-webgpu-profile`);
const llmCandidates = parseList(
process.env.BROWSER_SPEAK_WEBGPU_LLMS ??
"HuggingFaceTB/SmolLM2-135M-Instruct,onnx-community/granite-4.0-350m-ONNX-web,onnx-community/Qwen3-0.6B-ONNX",
);
const stackBase = {
device: "webgpu",
asr: process.env.BROWSER_SPEAK_ASR ?? "onnx-community/moonshine-base-ONNX",
voice: process.env.BROWSER_SPEAK_VOICE ?? "F2",
ttsSteps: Number(process.env.BROWSER_SPEAK_TTS_STEPS ?? 2),
vadSilenceMs: Number(process.env.BROWSER_SPEAK_VAD_SILENCE_MS ?? 480),
partialAsr: process.env.BROWSER_SPEAK_PARTIAL_ASR !== "false",
};
const loadTimeoutMs = Number(process.env.BROWSER_SPEAK_LOAD_TIMEOUT_MS ?? 600000);
const suiteTimeoutMs = Number(process.env.BROWSER_SPEAK_SUITE_TIMEOUT_MS ?? 900000);
const headless = process.env.BROWSER_SPEAK_HEADLESS !== "false";
const allowSoftwareWebgpu = process.env.BROWSER_SPEAK_ALLOW_SOFTWARE_WEBGPU === "true";
const protocolTimeoutMs = Number(process.env.BROWSER_SPEAK_CDP_TIMEOUT_MS ?? 60000);
const pollTimeoutMs = Number(process.env.BROWSER_SPEAK_CDP_POLL_TIMEOUT_MS ?? 5000);
async function main() {
await ensureServer();
await mkdir(dirname(resultPath), { recursive: true });
await rm(profileDir, { recursive: true, force: true });
const browser = launchBrowser(9343, profileDir);
const candidates = llmCandidates.map((llm) => ({ llm, status: "pending" }));
try {
const client = await connectToPage(9343, browser);
await waitForBenchApi(client);
const webgpu = await runPageTask(client, "window.browserSpeakBench.webgpuInfo()", {
label: "WebGPU probe",
timeoutMs: 30000,
});
const skipReason = webgpuSkipReason(webgpu);
if (skipReason) {
for (const candidate of candidates) {
candidate.status = "skipped";
candidate.reason = skipReason;
}
await writePayload({
skipped: true,
reason: skipReason,
webgpu,
config: benchmarkConfig(),
candidates,
results: [],
summary: null,
});
console.log(`Wrote WebGPU benchmark JSON: ${resultPath}`);
console.log(`Skipped WebGPU benchmark: ${skipReason}.`);
await client.closeBrowser();
return;
}
await runPageTask(client, "window.browserSpeakBench.clearResults()", { label: "clear results" });
console.log(`WebGPU adapter: ${formatAdapter(webgpu.adapter)}`);
for (const candidate of candidates) {
const stack = { ...stackBase, llm: candidate.llm };
const label = shortModelName(candidate.llm);
console.log(`Running WebGPU benchmark candidate: ${candidate.llm}`);
const startedAt = Date.now();
try {
const before = await runPageTask(client, "window.browserSpeakBench.state()", {
label: `${label} preflight`,
});
await runPageTask(client, `window.browserSpeakBench.loadStack(${JSON.stringify(stack)})`, {
label: `${label} model load`,
timeoutMs: loadTimeoutMs,
});
const snapshot = await runPageTask(client, "window.browserSpeakBench.runSuite()", {
label: `${label} suite`,
timeoutMs: suiteTimeoutMs,
});
const addedCount = Math.max(0, snapshot.results.length - before.results.length);
const rows = snapshot.results.slice(0, addedCount);
const errors = rows.filter((row) => row.error);
candidate.status = rows.length >= 5 && errors.length === 0 ? "complete" : "partial";
candidate.durationMs = Date.now() - startedAt;
candidate.rowCount = rows.length;
candidate.errorCount = errors.length;
candidate.summary = snapshot.summary.current ?? snapshot.summary.all;
candidate.stack = snapshot.stack;
if (errors.length > 0) candidate.errors = errors.map((row) => `${row.kind}: ${row.error}`);
} catch (error) {
candidate.status = "failed";
candidate.durationMs = Date.now() - startedAt;
candidate.error = error.message;
console.error(`${candidate.llm} failed: ${error.message}`);
} finally {
await runPageTask(client, "window.browserSpeakBench.stop()", {
label: `${label} stop`,
timeoutMs: 15000,
}).catch(() => {});
const state = await runPageTask(client, "window.browserSpeakBench.state()", {
label: `${label} state`,
timeoutMs: 15000,
}).catch(() => null);
if (state?.modelsLoaded || state?.modelsLoading) {
await runPageTask(client, "window.browserSpeakBench.unload()", {
label: `${label} unload`,
timeoutMs: 60000,
}).catch((error) => {
candidate.unloadError = error.message;
});
}
await writePayload(await exportPayload(client, webgpu, candidates));
}
}
const payload = await exportPayload(client, webgpu, candidates);
await writePayload(payload);
console.log(`Wrote WebGPU benchmark JSON: ${resultPath}`);
summarizeResults(payload);
await client.closeBrowser();
} finally {
await stopBrowser(browser, profileDir);
}
}
async function exportPayload(client, webgpu, candidates) {
const exportResult = await runPageTask(client, "window.browserSpeakBench.exportResults()", {
label: "export results",
timeoutMs: 30000,
}).catch(() => ({ summary: null, results: [] }));
return {
generatedAt: new Date().toISOString(),
sourceFingerprint: await sourceFingerprint(),
url,
webgpu,
config: benchmarkConfig(),
candidates,
summary: exportResult.summary,
results: exportResult.results,
};
}
function benchmarkConfig() {
return {
stackBase,
loadTimeoutMs,
suiteTimeoutMs,
headless,
allowSoftwareWebgpu,
protocolTimeoutMs,
pollTimeoutMs,
chrome,
extraChromeArgs: parseChromeArgs(),
};
}
async function writePayload(payload) {
await writeFile(
resultPath,
`${JSON.stringify(
{
generatedAt: new Date().toISOString(),
sourceFingerprint: await sourceFingerprint(),
url,
...payload,
},
null,
2,
)}\n`,
);
}
async function ensureServer() {
const response = await fetch(url).catch((error) => {
throw new Error(`Could not reach ${url}: ${error.message}`);
});
if (!response.ok) throw new Error(`${url} returned HTTP ${response.status}`);
}
async function waitForBenchApi(client) {
const deadline = Date.now() + 15000;
while (Date.now() < deadline) {
try {
if (await client.evaluate("Boolean(window.browserSpeakBench?.webgpuInfo)")) return;
} catch {
// The target may still be navigating and can destroy the execution context.
}
await sleep(100);
}
throw new Error("window.browserSpeakBench.webgpuInfo was not installed.");
}
async function runPageTask(client, expression, { label = "page task", timeoutMs = 30000 } = {}) {
const taskId = `task_${Date.now()}_${Math.random().toString(16).slice(2)}`;
await client.evaluate(`(() => {
const taskId = ${JSON.stringify(taskId)};
window.__browserSpeakHarnessTasks ||= {};
window.__browserSpeakHarnessTasks[taskId] = { done: false, label: ${JSON.stringify(label)} };
Promise.resolve(${expression})
.then((value) => {
window.__browserSpeakHarnessTasks[taskId] = { done: true, value };
})
.catch((error) => {
window.__browserSpeakHarnessTasks[taskId] = {
done: true,
error: error?.stack || error?.message || String(error),
};
});
return true;
})()`);
const deadline = Date.now() + timeoutMs;
let lastEvents = [];
let lastPollError = "";
while (Date.now() < deadline) {
let task = null;
try {
task = await client.evaluate(
`window.__browserSpeakHarnessTasks?.[${JSON.stringify(taskId)}] ?? null`,
pollTimeoutMs,
);
lastPollError = "";
} catch (error) {
const message = error.message ?? String(error);
if (message !== lastPollError) {
console.log(`${label}: waiting for page response (${message})`);
lastPollError = message;
}
await sleep(500);
continue;
}
if (task?.done) {
await client
.evaluate(`delete window.__browserSpeakHarnessTasks?.[${JSON.stringify(taskId)}]`, pollTimeoutMs)
.catch(() => {});
if (task.error) throw new Error(`${label} failed: ${task.error}`);
return task.value;
}
const snapshot = await client
.evaluate(`(() => {
const state = window.browserSpeakBench?.state?.();
return state ? {
modelsLoaded: state.modelsLoaded,
modelsLoading: state.modelsLoading,
activeBenchmark: state.activeBenchmark?.kind ?? null,
suiteRunning: state.suiteRunning,
events: state.events?.slice(0, 3) ?? [],
} : null;
})()`, pollTimeoutMs)
.catch(() => null);
const events = snapshot?.events ?? [];
if (events.join("\\n") !== lastEvents.join("\\n")) {
lastEvents = events;
if (events[0]) console.log(`${label}: ${events[0]}`);
}
await sleep(500);
}
throw new Error(`${label} timed out after ${(timeoutMs / 1000).toFixed(0)} seconds.`);
}
function launchBrowser(port, profileDir) {
const child = spawn(
chrome,
[
...(headless ? ["--headless=new"] : []),
"--no-sandbox",
"--disable-dev-shm-usage",
"--disable-background-networking",
"--disable-extensions",
"--no-default-browser-check",
"--no-first-run",
"--autoplay-policy=no-user-gesture-required",
"--enable-unsafe-webgpu",
"--ignore-gpu-blocklist",
`--remote-debugging-port=${port}`,
`--user-data-dir=${profileDir}`,
...parseChromeArgs(),
],
{ stdio: ["ignore", "pipe", "pipe"] },
);
child.browserLog = "";
const appendLog = (chunk) => {
child.browserLog = `${child.browserLog}${chunk}`;
if (child.browserLog.length > 8000) child.browserLog = child.browserLog.slice(-8000);
};
child.stdout.on("data", appendLog);
child.stderr.on("data", appendLog);
return child;
}
async function stopBrowser(child, profileDir) {
if (child.exitCode == null) child.kill("SIGTERM");
await new Promise((resolve) => {
child.once("exit", resolve);
setTimeout(resolve, 3000);
});
if (child.exitCode == null) child.kill("SIGKILL");
for (let attempt = 0; attempt < 5; attempt += 1) {
try {
await rm(profileDir, { recursive: true, force: true });
return;
} catch (error) {
if (attempt === 4) {
console.warn(`Could not remove ${profileDir}: ${error.message}`);
return;
}
await sleep(500);
}
}
}
async function connectToPage(port, child) {
const deadline = Date.now() + 60000;
let lastError = null;
while (Date.now() < deadline) {
if (child.exitCode != null) {
throw new Error(`Chrome exited before DevTools became available.\n${child.browserLog}`);
}
try {
const version = await fetch(`http://127.0.0.1:${port}/json/version`).then((response) => response.json());
if (version.webSocketDebuggerUrl) {
const page = await createPageTarget(port);
return new CdpClient(page.webSocketDebuggerUrl);
}
} catch (error) {
lastError = error;
}
await sleep(250);
}
throw new Error(
`Could not connect to Chrome DevTools on port ${port}: ${lastError?.message ?? "unknown error"}\n${child.browserLog}`,
);
}
async function createPageTarget(port) {
for (const method of ["PUT", "GET"]) {
const response = await fetch(`http://127.0.0.1:${port}/json/new?${encodeURIComponent(url)}`, {
method,
}).catch(() => null);
if (response?.ok) {
const target = await response.json();
if (target.webSocketDebuggerUrl) return target;
}
}
const targets = await fetch(`http://127.0.0.1:${port}/json`).then((response) => response.json());
const page = targets.find((target) => target.type === "page" && target.url === url);
if (page?.webSocketDebuggerUrl) return page;
throw new Error("Could not create or find a page target.");
}
class CdpClient {
constructor(webSocketUrl) {
this.nextId = 1;
this.pending = new Map();
this.socket = new WebSocket(webSocketUrl);
this.opened = new Promise((resolve, reject) => {
this.socket.onopen = resolve;
this.socket.onerror = reject;
this.socket.onmessage = (event) => this.onMessage(event);
});
}
onMessage(event) {
const message = JSON.parse(event.data);
if (!message.id || !this.pending.has(message.id)) return;
const { resolve: onResolve, reject } = this.pending.get(message.id);
this.pending.delete(message.id);
if (message.error) reject(new Error(message.error.message));
else onResolve(message.result);
}
async call(method, params = {}, timeoutMs = protocolTimeoutMs) {
await this.opened;
const id = this.nextId++;
this.socket.send(JSON.stringify({ id, method, params }));
return new Promise((resolvePromise, reject) => {
const timer = setTimeout(() => {
this.pending.delete(id);
reject(new Error(`${method} timed out after ${(timeoutMs / 1000).toFixed(0)} seconds.`));
}, timeoutMs);
this.pending.set(id, {
resolve: (value) => {
clearTimeout(timer);
resolvePromise(value);
},
reject: (error) => {
clearTimeout(timer);
reject(error);
},
});
});
}
async evaluate(expression, timeoutMs = protocolTimeoutMs) {
const result = await this.call("Runtime.evaluate", {
expression,
returnByValue: true,
}, timeoutMs);
if (result.exceptionDetails) throw new Error(formatException(result.exceptionDetails));
return result.result.value;
}
closeBrowser() {
try {
this.socket.send(JSON.stringify({ id: this.nextId++, method: "Browser.close", params: {} }));
} catch {
// The surrounding process cleanup handles already-closed targets.
}
try {
this.socket.close();
} catch {
// Ignore close races.
}
}
}
function formatException(exceptionDetails) {
const exception = exceptionDetails.exception;
return exception?.description ?? exception?.value ?? exceptionDetails.text ?? "Evaluation failed.";
}
function parseList(value) {
return value
.split(",")
.map((item) => item.trim())
.filter(Boolean);
}
function parseChromeArgs() {
const raw = process.env.BROWSER_SPEAK_CHROME_ARGS ?? "";
return (
raw
.match(/(?:[^\s"']+|"[^"]*"|'[^']*')+/g)
?.map((arg) => arg.replace(/^["']|["']$/g, "")) ?? []
);
}
function shortModelName(model) {
return model.split("/").at(-1) ?? model;
}
function formatAdapter(adapter) {
if (!adapter) return "unknown";
return [adapter.vendor, adapter.architecture, adapter.device, adapter.description].filter(Boolean).join(" / ") || "unknown";
}
function webgpuSkipReason(webgpu) {
if (!webgpu?.available) return "WebGPU unavailable";
if (allowSoftwareWebgpu) return "";
if (webgpu.softwareAdapter || isSoftwareAdapter(webgpu.adapter)) {
return `Software WebGPU adapter exposed (${formatAdapter(webgpu.adapter)})`;
}
return "";
}
function isSoftwareAdapter(adapter) {
const text = [adapter?.vendor, adapter?.architecture, adapter?.device, adapter?.description]
.filter(Boolean)
.join(" ")
.toLowerCase();
return /\b(swiftshader|llvmpipe|software rasterizer|software adapter|warp)\b/.test(text);
}
function summarizeResults(payload) {
for (const candidate of payload.candidates) {
const suffix = candidate.error ? ` (${candidate.error})` : "";
console.log(`${candidate.llm}: ${candidate.status}, ${candidate.rowCount ?? 0} rows${suffix}`);
}
const byStack = payload.summary?.byStack ?? {};
const stackCount = Object.keys(byStack).length;
console.log(`Exported ${payload.results.length} rows across ${stackCount} stack summaries.`);
}
function sleep(ms) {
return new Promise((resolve) => setTimeout(resolve, ms));
}
await main();