Update Qwen3-8B EAGLE3 example to online training pipeline

[cjluo-nv]

What

Updates tools/launcher/examples/Qwen/Qwen3-8B/hf_online_eagle3.yaml to use a streamlined online EAGLE3 training pipeline.

Changes

• Replaces the offline 4-step pipeline (data synthesis → hidden-state dump → offline training → benchmark) with a 2-step online pipeline:
  • task_0: Online training — train the EAGLE3 draft head against a prebuilt dataset
  • task_1: Benchmark — evaluate speculative decoding speedup via VLLM
• Uses a prebuilt conversation dataset (Speculative-Decoding-Dataset-v1-Qwen3-8B), so no data-synthesis step is needed.
• Adds a chat template (chat_template_train.jinja) for training.
• Removes now-unnecessary HOME / TORCHINDUCTOR_CACHE_DIR environment overrides.
• Updates header comments and usage examples to reference the online config.

🤖 Generated with Claude Code

[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: 3342ecef-08f8-4b28-a1b7-2351ad1b0ec3

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 chenjiel/qwen3-8b-online-eagle3

---

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