Revert "[Tiny] Enable Full Cuda Graph with Page size = 1" (#31062)

2026-07-13 14:23:44 -05:00
parent 805385414e
commit 86c59ac1aa
1 changed files with 0 additions and 3 deletions
@@ -1796,9 +1796,6 @@ def _fa4_page_constraint(view: Any) -> dict:
        # CUTLASS kernel aborts on at page_size>1. That path only works at
        # page_size==1, so skip the 128 auto-force for it and keep the default.
        and (view.speculative_eagle_topk or 0) <= 1
-        # The full prefill CUDA graph runs the FA backend at page_size==1 only
-        # (#27988), so skip the 128 auto-force for it and keep the default.
-        and view.cuda_graph_config.prefill.backend != Backend.FULL
    ):
        logger.warning(
            f"FA4 backend only supports page size 128 for non-MLA model architectures, changing page_size from {view.page_size} to 128."