Deprecate disable-mla (#5481)

python/sglang/srt/server_args.py

@@ -155,7 +155,6 @@ class ServerArgs:
     enable_nccl_nvls: bool = False
     disable_outlines_disk_cache: bool = False
     disable_custom_all_reduce: bool = False
-    disable_mla: bool = False
     enable_llama4_multimodal: Optional[bool] = None
     disable_overlap_schedule: bool = False
     enable_mixed_chunk: bool = False
@@ -975,11 +974,6 @@ class ServerArgs:
             action="store_true",
             help="Disable the custom all-reduce kernel and fall back to NCCL.",
         )
-        parser.add_argument(
-            "--disable-mla",
-            action="store_true",
-            help="Disable Multi-head Latent Attention (MLA) for DeepSeek V2/V3/R1 series models.",
-        )
         parser.add_argument(
             "--enable-llama4-multimodal",
             default=ServerArgs.enable_llama4_multimodal,