Remove stale vLLM compatibility warning for draft arch

[orestis-z]

Summary

• Remove the stale vLLM compatibility warning in create_transformer_layer_config that was missed by Change default drafter architecture from Llama to Qwen3 #620
• Both llama and qwen3 are supported in vLLM, so the warning is no longer needed

Test plan

• Run training with default args — no spurious warning emitted

🤖 Generated with Claude Code

[coderabbitai]

📝 Walkthrough

Walkthrough

In scripts/train.py, the create_transformer_layer_config function's draft_arch parameter default is changed from "llama" to "qwen3", and the warnings.warn call that previously fired when draft_arch != "llama" is removed.

Changes

draft_arch default and warning update

Layer / File(s)	Summary
Default and warning removal scripts/train.py	draft_arch parameter default changed from "llama" to "qwen3"; the conditional warnings.warn block for non-"llama" architectures is removed, while the validation against DRAFT_ARCH_CONFIGS is retained.

Possibly related PRs

• vllm-project/speculators#620: Directly modifies the same draft_arch default in scripts/train.py, switching from "llama" to "qwen3" in the same function.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title matches the main change by calling out removal of the stale vLLM compatibility warning.
Description check	✅ Passed	The description accurately summarizes the warning removal and notes the supported draft architectures.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @orestis-z.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@scripts/train.py`:
- Line 107: The default for draft_arch in the training flow is inconsistent
between the helper and the CLI/docs, so the new value is not actually used by
the main path. Update parse_args() in relation to the --draft-arch argument, the
related help text/docs, and the main() call site so they all agree with the
intended default, or keep the helper default in sync until those contracts are
changed together.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: d8246e1c-47cb-4da0-b859-336e62e17068

📥 Commits

Reviewing files that changed from the base of the PR and between c5e7b8a and 733e4f1.

📒 Files selected for processing (1)

• scripts/train.py

[orestis-z]

Closing — the upstream warning was already updated in a later commit to only warn for architectures other than llama/qwen3, which is valid. Nothing left to fix here.