This is a patch release on top of v0.18.0 to address a few issues:
- Change default SM100 MLA prefill backend back to TRT-LLM (#38562)
- Fix mock.patch resolution failure for standalone_compile.FakeTensorMode on Python <= 3.10 (#37158)
- Disable monolithic TRTLLM MoE for Renormalize routing #37605
- Pre-download missing FlashInfer headers in Docker build #38391
- Fix DeepGemm E8M0 accuracy degradation for Qwen3.5 FP8 on Blackwell (#38083)