[rhaiis] Add missing workload profiles 5-7 and composite benchmark presets

[hp2419]

Summary

• Sync workload profiles from model-furnace that were missing in forge:
  • profile5: 100k/1k ultra-long context (rates [1,2,5], 10 samples)
  • profile6: multi-turn with prefix caching (5 turns, prefix_tokens=512)
  • profile7: heavy heterogeneous (prompt 50-30k, output 20-8k)
• Add workload key preset mappings for profiles 5, 6, 7
• Add composite benchmark presets for common test matrices:
  • benchmark-standard: profile1 + profile2
  • benchmark-long-context: profile4 + profile5
  • benchmark-full-sweep: all profiles 1-7
  • benchmark-multi-turn: profile6
  • benchmark-heterogeneous: profile7

Test plan

• Verify profile5/6/7 workload definitions match model-furnace
• Verify composite presets resolve correctly via runtime_config
• Run ci-test preset to confirm no regressions

[openshift-ci]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by:
Once this PR has been reviewed and has the lgtm label, please assign tosokin for approval. For more information see the Code Review Process.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 9a2cd19d-a3cd-40fb-a3cb-e1a8b7faef18

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[hp2419]

cc @Harshith-umesh @albertoperdomo2 — could you review this when you get a chance?

[kpouget]

random question from a rhaiis-outsider: is there no way to give a meaningful name to the profiles?
profile6 seems to be multi-turn and profile7 is heterogeneous,

but are these number profiles mandatory? 🤔

[hp2419]

Great question! The numbered names (profile1–7) are inherited from model-furnace, where they're used in regression comparison (PROFILE_MAP), payload pattern matching
(_PROFILE_PATTERNS), and benchmark launch scripts. Renaming them in forge would break the alignment between the two systems and make cross-system result comparisons harder.

That said, the preset layer is exactly where we add meaningful names. The workload keys stay as profile1–profile7, but users interact through presets like benchmark-multi-turn,
benchmark-heterogeneous, benchmark-long-context, etc.

We could also consider adding aliases in workloads.yaml (e.g. multi-turn pointing to the same config as profile6) if we want meaningful names at the workload level too, without breaking the numbered convention.

[Harshith-umesh]

@hp2419 I'm closing this PR since @sacherukuri has already added this functionality in this PR: #173 a few days ago.