[Fix] Fix FlashInfer CUTLASS MoE for unquantized models and single-GPU, bump FlashInfer to 0.6.8#38215

Draft

askliar wants to merge 5 commits intovllm-project:mainfrom

askliar:feature/add_nemotronh_spark_support

Commits on Mar 26, 2026

[Fix] Update quant_scales initialization and correct comment typo in FlashInfer CUTLASS MoE implementation
Andrii Skliar
committed
[Update] Bump FlashInfer version to 0.6.7 in Dockerfile, versions.json, and requirements/cuda.txt
Andrii Skliar
committed
[Refactor] Remove use_ep parameter from unquantized backend selection in tests and implementation
Andrii Skliar
committed
[Update] Bump FlashInfer version to 0.6.8 in Dockerfile, versions.json, and requirements/cuda.txt
Andrii Skliar
committed
[Update] Enhance logging for FlashInfer CUTLASS MoE availability
Andrii Skliar
committed