DOCKER_README.md

Running RTEB with Docker

This document explains how to run the RTEB (Retrieval Embedding Benchmark) application using Docker.

Prerequisites

• Docker installed on your system
• Docker Compose installed on your system
• NVIDIA Docker runtime (only if you need GPU support)

Quick Start

1. Make sure Docker is running on your system.

2. Run the application with default settings:

   ./run_rteb.sh

   This will use the application's built-in defaults:

   • Data path: "data/"
   • Save path: "output/"
   • CPU mode (no GPUs)

3. Run with custom arguments:

   ./run_rteb.sh --gpus 2 --batch_size 32 --save_embds

Available Arguments

All arguments supported by the RTEB application can be passed directly to the Docker container. Here are some common ones:

• --gpus <num>: Number of GPUs to use (default: 0, requires NVIDIA Docker runtime)
• --cpus <num>: Number of CPUs to use (default: 1)
• --batch_size <num>: Batch size for encoding (default: 16)
• --data_path <path>: Path to the dataset (default: /app/data)
• --save_path <path>: Path to save output (default: /app/output)
• --save_embds: Save embeddings
• --load_embds: Load pre-computed embeddings
• --overwrite: Overwrite existing results

For a complete list of arguments, run:

./run_rteb.sh --help

Docker Configuration

The Docker setup includes:

1. A Docker image with all necessary dependencies
2. Volume mounts for data and output
3. Optional GPU support for accelerated processing (requires NVIDIA Docker runtime)
4. Memory limits to prevent out-of-memory errors

Customizing the Docker Environment

To modify the Docker environment:

1. Edit docker-compose.yml to change resource limits or volume mounts
2. Edit Dockerfile to modify the base image or installed dependencies
3. Edit docker-entrypoint.sh to change default arguments or startup behavior

After making changes, rebuild the Docker image:

sudo docker-compose build