FastAPI Model Serving for ML | DeployBytes 8 - Line by Line

Real MLOps, step by step. Part 8 serves the trained 7-encoder model over HTTP with FastAPI: a thin app on Uvicorn with /health and /predict, a lazy-loading singleton so the weights load once, a one-line Prometheus /metrics endpoint, and a container healthcheck and restart policy that keep the serving service self-healing. Why this series exists: This video is part of a series created to make central concepts easier to understand and hold onto. It was built with accessibility in mind, particularly for learners with learning differences, so the pace and style are kept simple and supportive. There may be a few mispronunciations in the spoken narration, and occasionally a small mistake in the content. These do not take away from the purpose of the series, which is simply to help the learning stick. Work with me:   / sandeep-grover-b3192a16   Productionizing a real 7-encoder multimodal Rakuten classifier (84,916 products, weighted-F1 0.9147), from git to Kubernetes. Chapters: 0:00 Why this series exists 0:29 Step 8 - Serving hardening: metrics, healthcheck, sacremoses 2:09 The shape of this step 4:30 What Step 8 is: one Uvicorn line, and where the request goes 6:09 The five endpoints 7:16 From a product title to a prdtypecode: featurize, standardize, 20-head bag 10:10 Provenance vs serving: the MLflow registry and the local file 11:26 Live run 12:38 8a - the metrics dependency: requirements-extra.txt and two new Dockerfile layers 14:00 8b - the missing tokenizer helper: sacremoses into requirements.txt, then rebuild 15:17 8c - /metrics in one line: the Instrumentator in api.py 16:34 8d - self-healing: restart policy and healthcheck on the serving service 18:57 8e - run it: the Uvicorn CMD, then /health and /predict from the host 20:04 8f - commit 20:42 curl the health endpoint on the container port 21:47 The shape of this step 22:51 curl the health endpoint through the proxy 24:23 The shape of this step 25:28 Walkthrough: src/serving/api.py 33:56 Walkthrough: docker-compose.yml 40:52 Walkthrough: docker/serving/Dockerfile 44:43 Walkthrough: docker/serving/requirements-extra.txt 45:53 Walkthrough: docker/serving/requirements.txt 49:15 Reality check: two errors, the slow first request, and /train 51:12 Design decisions to keep 52:58 Takeaway, and on to Step 9 Series stack (all 12 episodes): Python, cookiecutter-data-science, Git/GitHub, Docker, Docker Compose, PostgreSQL, MinIO (S3), MLflow, DVC, FastAPI, Prometheus, pytest, Nginx, Apache Airflow, Kubernetes, Argo Rollouts, PyTorch, Hugging Face Transformers. #MLOps #FastAPI #ModelServing #ModelDeployment #Prometheus #Docker #MachineLearning #Python #DevOps #MLEngineering