[hnyls2002]

• Include the selected DSA variant when checking exact-bucket graph availability with --disable-cuda-graph-padding; captured dense/sparse graphs previously fell back to eager.
• Keep padded admission free of length reads and preserve the independent LoRA key.
• Split the behavior fix from the graph-variant refactor in [Refactor] Generalize attention graph variants in the decode runner #38993, split from [Refactor] Separate V4.1 indexer capture options from candidate state #38962 for dsv4.1: remaining model and runtime integration #38798.
• Validation: one-off CPU checks reproduced the original fallback and passed after the fix locally and on lsyin-b300; pre-commit passed.

---

CI States

Latest PR Test (Base): 🚫 Run #34559599370
Latest PR Test (Extra): ❌ Run #34559599278
Latest PR Test (AMD ROCm 10): 🚫 Run #34559599340

[hnyls2002]

/rerun-test test/registered/unit/model_executor/runner/test_decode_cuda_graph_runner.py test/registered/unit/model_executor/runner_backend/test_full_cuda_graph_backend.py test/registered/unit/spec/test_eagle_draft_cuda_graph_runner.py

[github-actions]

Results for /rerun-test test/registered/unit/model_executor/runner/test_decode_cuda_graph_runner.py test/registered/unit/model_executor/runner_backend/test_full_cuda_graph_backend.py test/registered/unit/spec/test_eagle_draft_cuda_graph_runner.py:

🚀 ubuntu-latest (3 tests): ✅ View workflow run

cd test/ && python3 registered/unit/model_executor/runner/test_decode_cuda_graph_runner.py
cd test/ && python3 registered/unit/model_executor/runner_backend/test_full_cuda_graph_backend.py
cd test/ && python3 registered/unit/spec/test_eagle_draft_cuda_graph_runner.py