[Attention] Remove max cudagraph size limit of 992 (#27840)

Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com>

vllm/v1/attention/backends/flash_attn.py

@@ -244,13 +244,6 @@ class FlashAttentionMetadataBuilder(AttentionMetadataBuilder[FlashAttentionMetad
         self.max_cudagraph_size = self.compilation_config.max_cudagraph_capture_size
 
         if self.use_full_cuda_graph and self.aot_schedule:
-            if self.max_cudagraph_size > 992:
-                # This condition derives from FA3's internal heuristic.
-                # TODO(woosuk): Support larger cudagraph sizes.
-                raise ValueError(
-                    "Capture size larger than 992 is not supported for full cuda graph."
-                )
-
             self.scheduler_metadata = torch.zeros(
                 vllm_config.scheduler_config.max_num_seqs + 1,
                 dtype=torch.int32,

vllm/v1/attention/backends/mla/flashattn_mla.py

@@ -97,13 +97,6 @@ class FlashAttnMLAMetadataBuilder(MLACommonMetadataBuilder[FlashAttnMLAMetadata]
         self.max_cudagraph_size = self.compilation_config.max_cudagraph_capture_size
 
         if self.use_full_cuda_graph and self.fa_aot_schedule:
-            if self.max_cudagraph_size > 992:
-                # This condition derives from FA3's internal heuristic.
-                # TODO(woosuk): Support larger cudagraph sizes.
-                raise ValueError(
-                    "Capture size larger than 992 is not supported for full cuda graph."
-                )
-
             self.scheduler_metadata = torch.zeros(
                 vllm_config.scheduler_config.max_num_seqs + 1,
                 dtype=torch.int32,