TReqs Actions in Glaas
On this page
TReqs (treqs.ai) is a separate product for queueing and reviewing compute requests on your own infrastructure — think of it as pull requests for compute jobs. When TReqs schedules a run, the lineage gets auto-registered in GLaaS; from any GLaaS DAG, two actions connect you back into TReqs.
This page is relevant if your org uses TReqs (or you're considering it). If you only use roar and GLaaS, the GLaaS side works on its own — TReqs is an opt-in upgrade for orchestration and review.
View Training Request
On a DAG page, click View Training Request to open the request in TReqs.
If you have permission to view it on the TReqs side, you can see:
- Peer review and discussion
- Compute target activity, logs, and execution history
- Training-request timeline and updates
This gives you the collaboration and execution view of the same lineage context represented in GLaaS.
Run in TReqs
The Run in TReqs action clones a DAG context into TReqs as a new workflow-backed training request.
In practice: you launch work in your own TReqs org and project while preserving the original context for reproduction. Training requests behave like pull requests for compute — past work can be queued to run again with the exact same data, code, and environment, pending whatever approval gate your org has set.
If your org has no approval gate, the request runs immediately once submitted. If it does have one, the request waits for approval and you (or a reviewer) sees the full lineage context before the compute starts.
Generate a workflow from a session (roar workflow generate)
The two actions above start from a DAG already in GLaaS. roar workflow generate goes the other direction: it turns a local session into a TReqs workflow file you can commit, so a pipeline you ran ad hoc becomes something TReqs can discover and schedule.
roar workflow generate # active session -> .treqs/workflows/
roar workflow generate .treqs/workflows/train.yaml # choose the output path
roar workflow generate --session 8d7a1f2c --name baseline-train
It reads the session's recorded steps and writes workflow YAML compatible with TReqs workflow discovery, under .treqs/workflows/ by default.
| Flag | Effect |
|---|---|
--session <ref> | Which session to generate from — current (default) or a session-hash prefix (as shown by roar status / roar register). |
--name <name> | Override the generated workflow name. |
--force | Overwrite the output file if it already exists. |
The generated file is plain YAML — review it, commit it, and let TReqs pick it up. This is the on-ramp from "I traced a pipeline locally" to "TReqs can run it on demand."