Spaces:
Sleeping
Sleeping
File size: 3,138 Bytes
a668326 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 | [
{
"id": 0,
"start_sec": 0.0,
"end_sec": 7.0,
"text": "Welcome to this short introduction to MLOps, the practice of taking machine learning models to production.",
"language": "en"
},
{
"id": 1,
"start_sec": 7.0,
"end_sec": 16.0,
"text": "In this overview you'll see the full pipeline on the slides: data, training, serving, and monitoring.",
"language": "en"
},
{
"id": 2,
"start_sec": 16.0,
"end_sec": 25.0,
"text": "Everything starts with a good dataset. We collect, clean, and label our training data.",
"language": "en"
},
{
"id": 3,
"start_sec": 25.0,
"end_sec": 35.0,
"text": "On screen you can see the dataset summary, with the number of samples and the class balance.",
"language": "en"
},
{
"id": 4,
"start_sec": 35.0,
"end_sec": 45.0,
"text": "Next we move on to training. We fit a simple model and track its accuracy on a validation set.",
"language": "en"
},
{
"id": 5,
"start_sec": 45.0,
"end_sec": 55.0,
"text": "Here is the training curve. Notice how the validation loss flattens after a few epochs.",
"language": "en"
},
{
"id": 6,
"start_sec": 55.0,
"end_sec": 65.0,
"text": "Once training is done, we save the model and wrap it in an inference function.",
"language": "en"
},
{
"id": 7,
"start_sec": 65.0,
"end_sec": 75.0,
"text": "Now let's talk about serving. We expose the model through a REST API so other services can call it.",
"language": "en"
},
{
"id": 8,
"start_sec": 75.0,
"end_sec": 85.0,
"text": "The API has a predict endpoint that takes a JSON request and returns the model's prediction.",
"language": "en"
},
{
"id": 9,
"start_sec": 85.0,
"end_sec": 95.0,
"text": "How do we ship this to production reliably? We package everything into a Docker container.",
"language": "en"
},
{
"id": 10,
"start_sec": 95.0,
"end_sec": 105.0,
"text": "After deployment, monitoring becomes critical. We track latency, errors, and prediction quality.",
"language": "en"
},
{
"id": 11,
"start_sec": 105.0,
"end_sec": 115.0,
"text": "Let me show you the metrics dashboard, with p50 and p95 latency for each endpoint.",
"language": "en"
},
{
"id": 12,
"start_sec": 115.0,
"end_sec": 125.0,
"text": "We also add a health check endpoint so the platform can restart the service if it fails.",
"language": "en"
},
{
"id": 13,
"start_sec": 125.0,
"end_sec": 135.0,
"text": "A quick question: what happens when the data distribution changes over time? That is called drift.",
"language": "en"
},
{
"id": 14,
"start_sec": 135.0,
"end_sec": 145.0,
"text": "To summarize, MLOps connects data, training, serving, deployment, and monitoring into one pipeline.",
"language": "en"
},
{
"id": 15,
"start_sec": 145.0,
"end_sec": 150.0,
"text": "Thanks for watching this overview. In the next video we'll build the API step by step.",
"language": "en"
}
] |