How to Integrate Replit with TensorFlow

1# replit.nix — add system dependencies for Python TensorFlow
2# Edit this file in your Replit project if tensorflow-cpu fails to install
3{ pkgs }: {
4  deps = [
5    pkgs.python311
6    pkgs.python311Packages.pip
7    pkgs.stdenv.cc.cc.lib  # provides libstdc++.so.6
8    pkgs.zlib
9    pkgs.glib
10  ];
11}

1# Python: load MobileNetV2 for image classification
2import os
3os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'  # suppress TensorFlow startup messages
4
5import tensorflow as tf
6from tensorflow.keras.applications import MobileNetV2
7from tensorflow.keras.applications.mobilenet_v2 import preprocess_input, decode_predictions
8from tensorflow.keras.preprocessing import image
9import numpy as np
10
11print(f'TensorFlow version: {tf.__version__}')
12print('Loading MobileNetV2 model...')
13
14# Load pre-trained model (downloads ~14MB on first run, then cached)
15model = MobileNetV2(weights='imagenet', include_top=True)
16print('Model loaded successfully.')
17print(f'Input shape: {model.input_shape}')
18
19# Test with a dummy image
20test_input = np.random.random((1, 224, 224, 3)).astype('float32')
21test_input = preprocess_input(test_input)
22predictions = model.predict(test_input, verbose=0)
23top_preds = decode_predictions(predictions, top=3)[0]
24
25print('Test predictions (random input):')  
26for class_name, label, score in top_preds:
27    print(f'  {label}: {score:.3f}')
28print('Model inference working correctly.')

1# Flask prediction API for TensorFlow MobileNetV2
2import os
3os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
4
5from flask import Flask, request, jsonify
6import tensorflow as tf
7from tensorflow.keras.applications import MobileNetV2
8from tensorflow.keras.applications.mobilenet_v2 import preprocess_input, decode_predictions
9from tensorflow.keras.preprocessing import image
10import numpy as np
11import requests
12from PIL import Image
13from io import BytesIO
14
15app = Flask(__name__)
16
17# Load model once at startup
18print('Loading TensorFlow model...')
19model = MobileNetV2(weights='imagenet', include_top=True)
20model.compile()  # warm up the model
21print('Model ready.')
22
23def load_image_from_url(url):
24    """Download and preprocess image for MobileNetV2."""
25    response = requests.get(url, timeout=10)
26    img = Image.open(BytesIO(response.content)).convert('RGB')
27    img = img.resize((224, 224))
28    img_array = image.img_to_array(img)
29    img_array = np.expand_dims(img_array, axis=0)
30    return preprocess_input(img_array)
31
32@app.route('/predict', methods=['POST'])
33def predict():
34    data = request.get_json()
35    if not data or 'image_url' not in data:
36        return jsonify({'error': 'image_url required in request body'}), 400
37
38    try:
39        img_array = load_image_from_url(data['image_url'])
40        predictions = model.predict(img_array, verbose=0)
41        top_preds = decode_predictions(predictions, top=5)[0]
42
43        results = [
44            {'class': label, 'label': class_name, 'confidence': float(score)}
45            for class_name, label, score in top_preds
46        ]
47        return jsonify({'predictions': results})
48
49    except Exception as e:
50        return jsonify({'error': str(e)}), 500
51
52@app.route('/health')
53def health():
54    return jsonify({'status': 'ok', 'model': 'MobileNetV2'})
55
56if __name__ == '__main__':
57    app.run(host='0.0.0.0', port=3000)

1// Node.js TensorFlow.js image classification server
2// Install: npm install @tensorflow/tfjs-node @tensorflow-models/mobilenet @tensorflow/tfjs-backend-cpu
3const express = require('express');
4const tf = require('@tensorflow/tfjs-node');
5const mobilenet = require('@tensorflow-models/mobilenet');
6const https = require('https');
7const app = express();
8app.use(express.json());
9
10let model;
11
12async function loadModel() {
13  console.log('Loading MobileNet model...');
14  model = await mobilenet.load({ version: 2, alpha: 0.5 });
15  console.log('MobileNet model loaded and ready.');
16}
17
18app.post('/predict', async (req, res) => {
19  if (!model) return res.status(503).json({ error: 'Model not loaded yet' });
20  const { image_url } = req.body;
21  if (!image_url) return res.status(400).json({ error: 'image_url required' });
22
23  try {
24    // Download image and convert to tensor
25    const response = await fetch(image_url);
26    const buffer = Buffer.from(await response.arrayBuffer());
27    const tfImage = tf.node.decodeImage(buffer, 3);
28    const predictions = await model.classify(tfImage);
29    tfImage.dispose(); // free GPU memory
30    res.json({ predictions });
31  } catch (err) {
32    res.status(500).json({ error: err.message });
33  }
34});
35
36app.get('/health', (req, res) => res.json({ status: 'ok', model: 'MobileNet TFJS' }));
37
38loadModel().then(() => {
39  app.listen(3000, '0.0.0.0', () => console.log('TensorFlow.js server running on port 3000'));
40});

1# Python: model management with external storage fallback
2import os
3import boto3  # install: boto3
4import tensorflow as tf
5
6MODEL_DIR = '/tmp/model'
7S3_BUCKET = os.environ.get('MODEL_S3_BUCKET', '')
8S3_KEY = os.environ.get('MODEL_S3_KEY', 'mobilenet_v2.keras')
9
10def ensure_model_downloaded():
11    """Download model from S3 to /tmp if not already present."""
12    if os.path.exists(MODEL_DIR):
13        print(f'Model already in {MODEL_DIR}')
14        return
15
16    if S3_BUCKET:
17        print(f'Downloading model from S3: s3://{S3_BUCKET}/{S3_KEY}')
18        os.makedirs(MODEL_DIR, exist_ok=True)
19        s3 = boto3.client(
20            's3',
21            aws_access_key_id=os.environ['AWS_ACCESS_KEY_ID'],
22            aws_secret_access_key=os.environ['AWS_SECRET_ACCESS_KEY']
23        )
24        s3.download_file(S3_BUCKET, S3_KEY, f'{MODEL_DIR}/model.keras')
25        print('Model downloaded successfully.')
26    else:
27        # Fallback: use Keras pre-trained model
28        print('No S3 bucket configured — using Keras Applications model')
29
30def load_model():
31    ensure_model_downloaded()
32    model_path = f'{MODEL_DIR}/model.keras'
33    if os.path.exists(model_path):
34        return tf.keras.models.load_model(model_path)
35    else:
36        from tensorflow.keras.applications import MobileNetV2
37        return MobileNetV2(weights='imagenet')
38
39model = load_model()