|
<!doctype html>
|
|
<html>
|
|
<head>
|
|
<title>Efficiency Tests</title>
|
|
<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs/dist/tf.min.js" type="text/javascript"></script>
|
|
<script src="models.js" type="text/javascript"></script>
|
|
<script type="text/javascript">
|
|
|
|
var running = false;
|
|
var stop = false;
|
|
var timings = {};
|
|
|
|
models.sort()
|
|
|
|
function getMean(array) {
|
|
return array.reduce((a, b) => a + b) / array.length;
|
|
}
|
|
|
|
async function measureTimings() {
|
|
const button = document.getElementById("button");
|
|
const log = document.getElementById("log");
|
|
|
|
if (running) {
|
|
running = false;
|
|
button.innerHTML = "Stopping ...";
|
|
return;
|
|
}
|
|
|
|
running = true;
|
|
button.innerHTML = "Stop";
|
|
|
|
const x = tf.randomNormal([1, 128, 128, 1]);
|
|
timings = {};
|
|
|
|
for (const modelName of models) {
|
|
if (modelName in timings) continue;
|
|
|
|
tf.engine().startScope()
|
|
const modelUrl = modelName + '/model.json';
|
|
console.log('Loading: ' + modelUrl);
|
|
const model = await tf.loadGraphModel(modelUrl);
|
|
|
|
|
|
await model.predict(x);
|
|
|
|
const n = 100;
|
|
let times = [];
|
|
for (let i = 0; i < 5; ++i) {
|
|
const time = await tf.time(() => {
|
|
for (let i = 0; i < n; ++i) model.predict(x);
|
|
});
|
|
times.push(time);
|
|
}
|
|
tf.engine().endScope()
|
|
|
|
console.log(times);
|
|
|
|
timings[modelName] = times;
|
|
|
|
const t = times.map(x => x.wallMs / n);
|
|
const mean = getMean(t);
|
|
const std = Math.sqrt(getMean(t.map(x => (x - mean) ** 2)));
|
|
log.innerHTML += `<tr><td>${modelName}</td><td>${mean.toFixed(2)} ± ${std.toFixed(2)}</td>`;
|
|
|
|
if (!running) {
|
|
button.innerHTML = "Start";
|
|
break;
|
|
}
|
|
}
|
|
|
|
running = false;
|
|
button.innerHTML = "Start";
|
|
}
|
|
</script>
|
|
</head>
|
|
<body>
|
|
<button id="button" onclick="measureTimings()">Start</button>
|
|
<table id="log">
|
|
<tr><th>model</th><th>timing</th></tr>
|
|
</table>
|
|
</body>
|
|
</html> |
