Full Sweep Tests

[cquil11]

@yunzhoul-nv I see you have two full sweeps scheduled:

• https://github.com/InferenceMAX/InferenceMAX/actions/runs/20109570586
• https://github.com/InferenceMAX/InferenceMAX/actions/runs/20109575517

Do these need to be full sweeps? Or can they just be GB200 sweeps?

[yunzhoul-nv]

Hey @cquil11 , I have already temporarily disabled all other runners in my side branch so that these two full sweeps runs exclusively Dynamo + SGLang GB200. Here is the commit: InferenceMAX/InferenceMAX@c03076b.

[yunzhoul-nv]

We are using this temporary approach (modified full sweep to run only on GB200) to test our GB200 workflow before we consider the MR ready for review.

Previously I recall there was a GB200-e2e workflow that we can quickly test GB200-only configs, but I think this workflow is lost when we refactor the codebase?

Please let me know if there's a better way for us to do GB200 test only. Thanks!

[yunzhoul-nv]

Hey @cquil11 , I have reverted all changes to full sweeps (commit) and launched new sets of experiments using the e2e test workflow instead.

[cquil11]

Thank you @yunzhoul-nv , perhaps we should increase viz of docs for this
Sorry if it was confusing!