Student: Sam Touahri (300234041)

SEG4180 Lab 1 – Pretrained Model Dockerized Service

Overview

This project implements a REST API that serves a pretrained Hugging Face embedding model inside a Docker container.

The service exposes:

• GET /health – service status and model info
• POST /embed – returns embeddings for a list of texts
• POST /similarity – returns cosine similarity between two texts

The model used:

sentence-transformers/all-MiniLM-L6-v2

• Embedding dimension: 384
• Lightweight (~90MB model)
• CPU-friendly
• Suitable for semantic similarity and search tasks

---

Project Structure

lab1-model-service/

• src/
  • app.py
• tests/
  • smoke.sh
• Dockerfile
• requirements.txt
• README.md

---

API Endpoints

1. Health Check

GET /health

Example:

curl http://localhost:5000/health

Response:

{ "status": "ok", "model": "sentence-transformers/all-MiniLM-L6-v2", "embedding_dimension": 384 }

---

2. Embed Texts

POST /embed

Request body:

{ "texts": ["pavement", "sidewalk", "bottle", "cup"] }

Example:

curl -X POST http://localhost:5000/embed
-H "Content-Type: application/json"
-d '{"texts":["pavement","sidewalk","bottle","cup"]}'

Response:

{ "model": "sentence-transformers/all-MiniLM-L6-v2", "dimension": 384, "embeddings": [[...], [...], [...], [...]] }

---

3. Similarity

POST /similarity

Request body:

{ "text_a": "pavement", "text_b": "sidewalk" }

Example:

curl -X POST http://localhost:5000/similarity
-H "Content-Type: application/json"
-d '{"text_a":"pavement","text_b":"sidewalk"}'

Response:

{ "model": "sentence-transformers/all-MiniLM-L6-v2", "dimension": 384, "similarity": 0.68, "text_a": "pavement", "text_b": "sidewalk" }

---

Local Development (Without Docker)

1. Create Environment

conda create -n seg4180-lab1 python=3.11 -y
conda activate seg4180-lab1

2. Install Dependencies

pip install -r requirements.txt

3. Run Application

python src/app.py

Server runs at:

http://localhost:5000

4. Run Smoke Test

./tests/smoke.sh http://localhost:5000

---

Docker Instructions

1. Build Image

docker build -t lab1-embed-service:latest .

2. Run Container

docker run --rm -p 5000:5000 lab1-embed-service:latest

3. Test Container

./tests/smoke.sh http://localhost:5000

---

Docker Hub

Docker Hub Link:

https://hub.docker.com/repository/docker/otoua046/seg4180-embed-service/general

Image repository:

otoua046/seg4180-embed-service

Pull image:

docker pull otoua046/seg4180-embed-service:latest

Run from Docker Hub image:

docker run --rm -p 5000:5000 otoua046/seg4180-embed-service:latest

---

Technology Stack

• Python 3.11
• Flask
• Waitress (WSGI server)
• sentence-transformers
• Torch
• Docker

---

Notes

• Model loads once at container startup.
• Embeddings are normalized, so similarity is computed using dot product (cosine similarity).
• Image built for linux/arm64.
• Service is production-ready using Waitress instead of Flask development server.

---

Submission Includes

• Application code (src/app.py)
• Dockerfile
• requirements.txt
• Smoke test script
• Screenshots demonstrating:
  • Running container
  • Working API calls
  • Docker Hub repository page
• Docker Hub link