I've struggling to find where a memory leak occurs in this file. This file is exported as an Event Listener. For context, I have 92 shards (meaning 92 of these listeners) running. I import the model from outside of this file so it's only loaded once per shard occurrence (stable 75 tensors in memory). However, after a few minutes, all the RAM on my computer is consumed (the function inside the file is called a dozen or so times per second). Have I overlooked any place which may cause this memory leak?
const use = require(`#tensorflow-models/universal-sentence-encoder`);
const tf = require(`#tensorflow/tfjs-node`);
const run = async (input, model) => {
const useObj = await use.load();
const encodings = [ await useObj.tokenizer.encode(input) ];
const indicesArr = encodings.map(function (arr, i) { return arr.map(function (d, index) { return [i, index]; }); });
var flattenedIndicesArr = [];
for (i = 0; i < indicesArr.length; i++) {
flattenedIndicesArr = flattenedIndicesArr.concat(indicesArr[i]);
}
const indices = tf.tensor2d(flattenedIndicesArr, [flattenedIndicesArr.length, 2], 'int32')
const value = tf.tensor1d(tf.util.flatten([ encodings ]), 'int32')
const prediction = await model.executeAsync({ Placeholder_1: indices, Placeholder: value });
const classes = [ 'Identity Attack', 'Insult', 'Obscene', 'Severe Toxicity', 'Sexual Explicit', 'Threat', 'Toxicity' ]
let finArr = [];
let finMsg = `Input: ${input}, `;
for (i = 0; i < prediction.length; i++) {
const sorted = tf.topk(prediction[i], 2);
const predictions = [ sorted.values.arraySync(), sorted.indices.arraySync() ];
const percentage = (predictions[0][0][0]*100).toFixed(2);
if (predictions[1][0][0] == 1) {
finArr.push(`${classes[i]} (${percentage}%)`);
}
tf.dispose([ sorted, predictions ]);
}
for (i = 0; i < finArr.length; i++) {
finMsg+=`${finArr[i]}, `;
}
tf.dispose([ prediction, indices, value, useObj ]);
console.log(finMsg);
console.log(tf.memory());
};
const main = async (message, client, Discord, model) => {
if (message.author.bot) return;
const input = message.content;
await run(input, model);
};
module.exports = {
event: 'messageCreate',
run: async (message, client, Discord, model) => {
await main(message, client, Discord, model);
},
};
to start with, you say this runs multiple times - so why are you loading model again and again? and disposing model is tricky, big chance that's part of your memory leak.
move const useObj = await use.load() outside of run loop and don't dispose it until you're done with all of the runs.
Related
I'm using Tensorflow js in react native and I'm getting the correct predictions for my model but it takes a lot of time to give results. For eg I'm using a custom model created by me in teachable machine by Google. But the .datasync() takes time approx. 1 second whole to give results. This causes a physical lag in the camera I want to get results instantly. This is my code below: -
<TensorCamera
style={styles.camera}
flashMode={Camera.Constants.FlashMode.off}
type={Camera.Constants.Type.back}
resizeWidth={224}
resizeHeight={224}
resizeDepth={3}
onReady={handleCameraStream}
autorender={true}
/>
//
const handleCameraStream = (imageAsTensors) => {
try {
} catch (e) {
// console.log("Tensor 1 not found!");
}
const loop = async () => {
// && detected == true
if (model !== null) {
if (frameCount % makePredictionsEveryNFrames === 0) {
const imageTensor = imageAsTensors.next().value;
await getPrediction(imageTensor);
// .catch(e => console.log(e));
}
}
frameCount += 1;
frameCount = frameCount % makePredictionsEveryNFrames;
requestAnimationFrameId = requestAnimationFrame(loop);
};
loop();
//loop infinitely to constantly make predictions
};
//
const getPrediction = async (tensor) => {
// if (!videoLink) {
if (!tensor) {
console.log("Tensor not found!");
return;
}
//
const imageData2 = tensor.resizeBilinear([224, 224]);
// tf.image.resizeBilinear(tensor, [224, 224]);
const normalized = imageData2.cast("float32").div(127.5).sub(1);
const final = tf.expandDims(normalized, 0);
//
console.time();
const prediction = model.predict(final).dataSync();
console.timeEnd();
console.log("Predictions:", prediction);
}
I heard about using .data() instead of .datasync() but I don't know how to implement .data() in my current code. please help.
predict is what takes time - and that is really up to your model
maybe it can run faster on different backend (no idea which backend you're using, default for browsers would be webgl), but in reality it is what it is without rearchitecting the model items.
datasync simply downloads results from wherever tensors are (e.g. in gpu vram) to your variable in js.
yes, you could use data instead which is an async call, but difference is couple of ms at best - its not going to speed up model execution at all.
btw, you're not releasing tensors anywhere - your application has some serious memory leaks.
I am processing an audio buffer with an OfflineAudioContext with the following node layout:
[AudioBufferSourceNode] -> [AnalyserNode] -> [OfflineAudioContext]
This works very good on Chrome (106.0.5249.119) but on Safari 16 (17614.1.25.9.10, 17614) each time I run the analysis takes longer and longer. Both running on macOS.
What's curious is that I must quit Safari to "reset" the processing time.
I guess there's a memory leak?
Is there anything that I'm doing wrong in the JavaScript code that would cause Safari to not garbage collect?
async function processFrequencyData(
audioBuffer,
options
) {
const {
fps,
numberOfSamples,
maxDecibels,
minDecibels,
smoothingTimeConstant,
} = options;
const frameFrequencies = [];
const oc = new OfflineAudioContext({
length: audioBuffer.length,
sampleRate: audioBuffer.sampleRate,
numberOfChannels: audioBuffer.numberOfChannels,
});
const lengthInMillis = 1000 * (audioBuffer.length / audioBuffer.sampleRate);
const source = new AudioBufferSourceNode(oc);
source.buffer = audioBuffer;
const az = new AnalyserNode(oc, {
fftSize: numberOfSamples * 2,
smoothingTimeConstant,
minDecibels,
maxDecibels,
});
source.connect(az).connect(oc.destination);
const msPerFrame = 1000 / fps;
let currentFrame = 0;
function process() {
const frequencies = new Uint8Array(az.frequencyBinCount);
az.getByteFrequencyData(frequencies);
// const times = new number[](az.frequencyBinCount);
// az.getByteTimeDomainData(times);
frameFrequencies[currentFrame] = frequencies;
const nextTime = (currentFrame + 1) * msPerFrame;
if (nextTime < lengthInMillis) {
currentFrame++;
const nextTimeSeconds = (currentFrame * msPerFrame) / 1000;
oc.suspend(nextTimeSeconds).then(process);
}
oc.resume();
}
oc.suspend(0).then(process);
source.start(0);
await oc.startRendering();
return frameFrequencies;
}
const buttonsDiv = document.createElement('div');
document.body.appendChild(buttonsDiv);
const initButton = document.createElement('button');
initButton.onclick = init;
initButton.innerHTML = 'Load audio'
buttonsDiv.appendChild(initButton);
const processButton = document.createElement('button');
processButton.disabled = true;
processButton.innerHTML = 'Process'
buttonsDiv.appendChild(processButton);
const resultElement = document.createElement('pre');
document.body.appendChild(resultElement)
async function init() {
initButton.disabled = true;
resultElement.innerText += 'Loading audio... ';
const audioContext = new AudioContext();
const arrayBuffer = await fetch('https://gist.githubusercontent.com/marcusstenbeck/da36a5fc2eeeba14ae9f984a580db1da/raw/84c53582d3936ac78625a31029022c8fdb734b2a/base64audio.txt').then(r => r.text()).then(fetch).then(r => r.arrayBuffer())
resultElement.innerText += 'finished.';
resultElement.innerText += '\nDecoding audio... ';
const audioBuffer = await audioContext.decodeAudioData(arrayBuffer);
resultElement.innerText += 'finished.';
processButton.onclick = async () => {
processButton.disabled = true;
resultElement.innerText += '\nStart processing... ';
const t0 = Date.now();
await processFrequencyData(audioBuffer, {
fps: 30,
numberOfSamples: 2 ** 13,
maxDecibels: -25,
minDecibels: -70,
smoothingTimeConstant: 0.2,
});
resultElement.innerText += `finished in ${Date.now() - t0} ms`;
processButton.disabled = false;
};
processButton.disabled = false;
}
I guess this is really a bug in Safari. I'm able to reproduce it by rendering an OfflineAudioContext without any nodes. As soon as I use suspend()/resume() every invocation takes a little longer.
I'm only speculating here but I think it's possible that there is some internal mechanism which tries to prevent the rapid back and forth between the audio thread and the main thread. It almost feels like one of those login forms which takes a bit longer to validate the password every time you try.
Anyway I think you can avoid using suspend()/resume() for your particular use case. It should be possible to create an OfflineAudioContext for each of the slices instead. In order to get the same effect you would only render the particular slice with each OfflineAudioContext.
const currentTime = 0;
while (currentTime < duration) {
const offlineAudioContext = new OfflineAudioContext({
length: LENGTH_OF_ONE_SLICE,
sampleRate
});
const audioBufferSourceNode = new AudioBufferSourceNode(
offlineAudioContext,
{
buffer
}
);
const analyserNode = new AnalyserNode(offlineAudioContext);
audioBufferSourceNode.start(0, currentTime);
audioBufferSourceNode
.connect(analyserNode)
.connect(offlineAudioContext.destination);
await offlineAudioContext.startRendering();
const frequencies = new Uint8Array(analyserNode.frequencyBinCount);
analyserNode.getByteFrequencyData(frequencies);
// do something with the frequencies ...
currentTime += LENGTH_OF_ONE_SLICE * sampleRate;
}
I think the only thing missing would be the smoothing since each of those slices will have it's own AnalyserNode.
const [Datalist,setDatalist] = useState([]);
useEffect(() => {
axios.get( 'http://0.0.0.0:8000/api/v1/questions/history/1')
.then(response => {
const questions = response.data;
const datalist = [];
for (let i = 0; i < questions.length - 1; i++) {
const data = new Object();
data.isExpanded = false;
data.question_id = questions[i].id;
data.question = questions[i].content;
data.type = questions[i].type;
data.commentType = questions[i].comment_type;
data.answer = [];
datalist.push(data);
}
setDatalist(datalist);
});
},[]);
I have three questions in my database currently. The for loop should be iterating through 0 to 2, however, it is only iterating twice.
And I'm also having problems putting the data into Datalist.
Anybody know where the issue is??
Thanks in advance!!
Change your for loop to this:
for (let i = 0; i < questions.length; i++)
Since you are iterating over each question you receive, you could use the map-method (if your environment supports ES6-Syntax - but since you're using react, it most likely dooes).
From the MDN Docs:
The map() method creates a new array populated with the results of calling a provided function on every element in the calling array.
With map, your code could look like this:
(Also note the removal of const data = new Object();. you can initialize an object and assign its properties/values at the same time)
const [Datalist,setDatalist] = useState([]);
useEffect(() => {
axios.get( 'http://0.0.0.0:8000/api/v1/questions/history/1')
.then(response => {
const questions = response.data;
const datalist = questions.map(question => {
return {
isExpanded: false;
question_id: question.id;
question: question.content;
type: question.type;
commentType: question.comment_type;
answer: [];
};
});
setDatalist(datalist);
});
},[]);
I have a vue function call which is triggered when selecting a radio button but it seems that my code inside my $nextTick is running before my previous line of code is finished. I don't want to use setTimout as I don't know how fast the user connection speed is.
findOrderer() {
axios.post('/MY/ENDPOINT')
.then((response) => {
this.orderers = response.data.accounts;
console.log('FIND_ORDER', this.orderers)
...OTHER_CODE
}
rbSelected(value) {
this.findOrderer();
this.newOrderList = [];
this.$nextTick(() => {
for (var i = 0, length = this.orderers.length; i < length; i++) {
console.log('FOR')
if (value.srcElement.value === this.orderers[i].accountType) {
console.log('IF')
this.newOrderList.push(this.orderers[i]);
}
}
this.$nextTick(() => {
this.orderers = [];
this.orderers = this.newOrderList;
console.log('orderers',this.orderers)
})
})
}
Looking at the console log the 'FINE_ORDERER' console.log is inside the 'findOrderer' function call so I would have expected this to be on top or am I miss using the $nextTick
That's expected, since findOrderer() contains asynchronous code. An easy way is to simply return the promise from the method, and then await it instead of waiting for next tick:
findOrderer() {
return axios.post('/MY/ENDPOINT')
.then((response) => {
this.orderers = response.data.accounts;
console.log('FIND_ORDER', this.orderers);
});
},
rbSelected: async function(value) {
// Wait for async operation to complete first!
await this.findOrderer();
this.newOrderList = [];
for (var i = 0, length = this.orderers.length; i < length; i++) {
console.log('FOR')
if (value.srcElement.value === this.orderers[i].accountType) {
console.log('IF')
this.newOrderList.push(this.orderers[i]);
}
}
this.orderers = [];
this.orderers = this.newOrderList;
console.log('orderers',this.orderers)
}
I'm trying another solution to my problem:
The thing is: im rendering a Sketch component with a background image and sketching over it
onReady = async () => {
const { layoutWidth, layoutHeight, points } = this.state;
this.sketch.graphics = new PIXI.Graphics();
const linesStored = this.props.screenProps.getSketchLines();
if (this.sketch.stage) {
if (layoutWidth && layoutHeight) {
const background = await PIXI.Sprite.fromExpoAsync(this.props.image);
background.width = layoutWidth * scaleR;
background.height = layoutHeight * scaleR;
this.sketch.stage.addChild(background);
this.sketch.renderer._update();
}
if (linesStored) {
for(let i = 0; i < linesStored.length; i++) {
this.sketch.stage.addChild(linesStored[i])
this.sketch.renderer._update();
}
}
}
};
this lineStored variable is returning a data i've saved onChange:
onChangeAsync = async (param) => {
const { uri } = await this.sketch.takeSnapshotAsync();
this.setState({
image: { uri },
showSketch: false,
});
if (this.sketch.stage.children.length > 0) {
this.props.screenProps.storeSketchOnState(this.sketch.stage.children);
}
this.props.callBackOnChange({
image: uri,
changeAction: this.state.onChangeAction,
startSketch: this.startSketch,
undoSketch: this.undoSketch,
});
};
storeSketchOnState saves this.sketch.stage.children; so when i change screen and back to the screen my Sketch component in being rendered, i can retrieve the sketch.stage.children from App.js state and apply to Sketch component to persist the sketching i was doing before
i'm trying to apply the retrieved data like this
if (linesStored) {
for(let i = 0; i < linesStored.length; i++) {
this.sketch.stage.addChild(linesStored[i])
this.sketch.renderer._update();
}
}
```
but it is not working =(