Deprecate global_server_args_dict (#11331)

python/sglang/srt/layers/attention/flashattention_backend.py

@@ -11,8 +11,8 @@ import triton.language as tl
 from sglang.srt.configs.model_config import AttentionArch
 from sglang.srt.layers.attention.base_attn_backend import AttentionBackend
 from sglang.srt.layers.radix_attention import AttentionType
-from sglang.srt.managers.schedule_batch import global_server_args_dict
 from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMode
+from sglang.srt.server_args import get_global_server_args
 from sglang.srt.speculative.spec_info import SpecInput
 
 if TYPE_CHECKING:
@@ -830,7 +830,7 @@ class FlashAttentionBackend(AttentionBackend):
             ):
                 # Do multi-head attention with chunked prefix cache
                 if forward_batch.attn_attend_prefix_cache:
-                    assert not global_server_args_dict["disable_chunked_prefix_cache"]
+                    assert not get_global_server_args().disable_chunked_prefix_cache
                     # MHA for chunked prefix kv cache when running model with MLA
                     assert forward_batch.prefix_chunk_idx is not None
                     assert forward_batch.prefix_chunk_cu_seq_lens is not None