Demo Video

A narrated walkthrough, in under four minutes: the command line, a method integration, a GPU run, the web dashboard, and a replayed experiment.

Chapters

Select a time to jump to that chapter.

Time	Chapter	What it shows
0:00	Introduction	What GraFlag is for.
0:14	How it works	The client, the Docker Swarm cluster, one service per experiment, and the shared NFS store.
0:31	Command line	graflag status, then the method and dataset catalogues.
0:54	Integrating a method	The files that make up RARE's integration, its .env pinned to the authors' commit, and the contract tests every method definition passes.
1:18	Running a method	graflag run on a GPU worker: the method's own AUROC, the runner's time and memory measurements, the service removed, then graflag evaluate.
2:04	Web dashboard	Launching AnoGraph on ISCX, evaluating it, and opening its plots.
2:58	Reproducing a run	Replaying a saved service_config.json on another worker, then evaluating the replay: the same AUC-ROC as the original.
3:30	Getting started	pip install graflag.

About the recording

The video was recorded on the development cluster that graflag devcluster deploys: a manager and four workers, running as Docker containers on one workstation and sharing its GPU, an NVIDIA RTX A2000 (12 GB). Swarm schedules them as five separate nodes, so the worker names in the logs are the ones a physical cluster would show.

Every command in the terminal scenes was really run, and the output on screen is exactly what it printed. Long pauses are shortened, and the one sped-up stretch, RARE’s search phase, is labelled on screen. The dashboard scene plays in real time.

Captions are available from the player’s captions menu.