Update run-e2e skill fixing inconsistencies for the Qwen suite

[ronenkat]

PLEASE FILL IN THE PR DESCRIPTION AND MAKE SURE THE CHECKLIST ITEMS HAVE BEEN CONSIDERED.

Purpose

1. Correct the skill's Scope section: the four gate scenarios and the 2A2F
   topology belong to DeepSeek-V2-Lite only; the Qwen3 MoE and Qwen3.6 suites
   are baseline-graph plus three 2A1F scenarios, which use the first two
   devices for Attention and the third for FFN and leave the fourth idle.
   Apply the same correction to the opening paragraph of
   tests/e2e/README.md.
2. Add HF_ENDPOINT to the skill's prerequisite checks and to its environment
   reference table, with the reason: gsm8k.py defaults the lm-eval child to
   hf-mirror.com, so an unreachable mirror is not visible until GSM8K starts,
   with the servers already up.

Issue

• Related issue(s): [Doc]: run-e2e skill misstates the Qwen suite topology and omits the HF_ENDPOINT prerequisite #305
• Closing keyword, only if fully resolved: Closes [Doc]: run-e2e skill misstates the Qwen suite topology and omits the HF_ENDPOINT prerequisite #305

Scope

• In scope: Update run-e2e skill and e2e readme
• Out of scope:

Implementation Notes

None

Test Plan

N/A

Test Result

N/A

Docs Impact

• Files updated:
• tests/e2e/README.md
• .agents/skills/run-e2e/SKILL.md
• If none, reason:

---

Essential PR Checklist

• [ X ] Purpose is clear and linked to public context when possible.
• [ X ] Scope is bounded.
• Compatibility with vLLM v0.26.0 is considered.
• [ X ] No changes are made to the vLLM source checkout.
• Plugin-owned classes or explicit dotted class paths are preferred over monkey patches.
• Any compat shim or monkey patch is isolated, idempotent, version-guarded, documented, and tested.
• Imports remain CPU-safe; CUDA-heavy work is delayed or GPU-gated.
• Validation evidence is included, including skipped GPU tests when applicable.
• [ X ] Documentation impact is stated.

[jiangkuaixue123]

The topology correction matches the Qwen suites, and the HF_ENDPOINT guidance matches gsm8k.py. Two remaining passages should be aligned with the corrected scope.

[jiangkuaixue123]

Please also update the “Local 2A1F cases” section below (lines 94–98). It still says all 2 Attention + 1 FFN scenarios are local-only and that CI gates do not run them, which contradicts the Qwen gate scope clarified here. Qualify both the heading and that sentence as DeepSeek-V2-Lite-specific, e.g. “The DeepSeek-V2-Lite 2A1F scenarios are local-only; its CI gate selects the 2A2F scenarios.”

[jiangkuaixue123]

Please also align the device-order paragraph below (lines 89–90), which still calls these “The local 2A1F scenarios.” Since this skill covers Qwen suites whose gate scenarios use this same topology, remove “local” there or explicitly qualify it as DeepSeek-V2-Lite-only. This will keep the execution guidance consistent with the corrected scope here.

[ronenkat]

The topology correction matches the Qwen suites, and the HF_ENDPOINT guidance matches gsm8k.py. Two remaining passages should be aligned with the corrected scope.

Thank you for the feedback. Text updated.