Use Cases

Where did this model come from?

Goal: start from a model artifact and understand how it was produced.

1. Locate the model artifact hash (from a prior roar register, from roar show, or from the GLaaS website).
2. On glaas.ai, paste the hash into the search bar.
3. On the artifact page:
   • review Associated Jobs
   • click into the job(s) to see inputs/outputs
   • click into the Originating Session to see the broader context

You can do the same locally:

roar show <model-hash>

This answers: which code, parameters, data, and upstream steps produced this artifact?

What did I train this on?

Goal: trace from an output artifact back to the input dataset(s).

1. Look up the model artifact on glaas.ai.
2. Click the producing job in Associated Jobs.
3. Follow input artifact links upstream until you reach:
   • the dataset artifact(s), or
   • external dataset references (e.g., downloaded shards)

This yields a concrete, hash-addressed answer rather than “whatever file name we used last week.”

I want to make that.

Goal: reproduce an artifact someone else produced.

1. Find the artifact hash on glaas.ai (or get it from a teammate).
2. Click “Reproduce with roar” (or copy the hash).
3. Run:

roar reproduce <artifact-hash>

4. Inspect the recipe DAG, then execute steps as needed.

This is especially useful for "make that exact thing again" workflows.