[Chore]: Update Dflash recipes to use dpace

[h-guo18]

What does this PR do?

Type of change: Chore (recipe update)

Makes D-PACE the position-loss weighting for all DFlash recipes, following #1724 (which set dflash_loss_objective=dpace as the config default). The recipes still carried dflash_loss_decay_factor, which is ignored once the objective defaults to dpace (and emits a runtime warning). This PR removes those now-dead overrides so the recipes inherit the dpace default, and documents how to opt back into static decay.

• modelopt_recipes/general/speculative_decoding/dflash.yaml: drop dflash_loss_decay_factor: 4.0, surface dflash_dpace_alpha: 0.5, and comment how to switch back to decay.
• 8 tools/launcher/examples/** DFlash recipes (Qwen3-8B / Qwen3-30B-A3B / Qwen3.5-4B / MiniMax-M2.7 / gpt-oss-20b): remove the dead dflash.dflash_loss_decay_factor=N override.

Domino recipes are intentionally left unchanged — Domino uses its own base/final dual loss and does not implement the dpace objective.

Usage

# DFlash recipes now default to D-PACE; no setting needed. To use static decay instead:
dflash:
  dflash_loss_objective: decay
  dflash_loss_decay_factor: 7   # gamma per block_size (7 for 16, 5 for 10, 4 for 8)

Testing

Recipe/config-only change; no code logic. The dpace default and config validation are covered by the unit tests in #1724 (tests/unit/torch/speculative/plugins/test_hf_dflash.py).

Before your PR is "Ready for review"

• Is this change backward compatible?: ❌ — these DFlash recipes switch from static decay to D-PACE weighting. Intended; opt back in via dflash_loss_objective: decay.
• If you copied code from any other sources or added a new PIP dependency, did you follow guidance in CONTRIBUTING.md: N/A
• Did you write any new necessary tests?: N/A (recipe-only; logic tested in [Feat]:Support DPace #1724)
• Did you update Changelog?: N/A (D-PACE default is recorded in [Feat]:Support DPace #1724)
• Did you get Claude approval on this PR?: N/A

Additional Information

Stacked on #1724 — set the PR base to haoguo/dpace so the diff shows only the recipe changes, and merge after #1724 (depends on its dflash_dpace_alpha config field).

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 74649f53-36a3-467d-a0cb-79f21fa6d95b

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch haoguo/udpate-dflash-recipe

---

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

[github-actions]

PR Preview Action v1.8.1
🚀 View preview at https://NVIDIA.github.io/Model-Optimizer/pr-preview/pr-1775/
Built to branch gh-pages at 2026-06-20 00:05 UTC. Preview will be ready when the GitHub Pages deployment is complete.

[codecov]

Codecov Report

❌ Patch coverage is 92.30769% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 76.64%. Comparing base (93dd08f) to head (a81526c).
⚠️ Report is 2 commits behind head on main.

Files with missing lines	Patch %	Lines
modelopt/torch/speculative/plugins/hf_dflash.py	92.30%	2 Missing ⚠️
modelopt/torch/speculative/dflash/dflash_model.py	83.33%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1775      +/-   ##
==========================================
- Coverage   76.92%   76.64%   -0.29%     
==========================================
  Files         511      511              
  Lines       56360    56395      +35     
==========================================
- Hits        43356    43225     -131     
- Misses      13004    13170     +166     

Flag	Coverage Δ
unit	54.35% <64.10%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.