Revert "[Fix] DeepSeek EP accuracy issue on B200 GPUs (#9946)" (#9955)

python/sglang/srt/server_args.py

@@ -268,7 +268,6 @@ class ServerArgs:
     flashinfer_mxfp4_moe_precision: Literal["default", "bf16"] = "default"
     enable_flashinfer_allreduce_fusion: bool = False
     deepep_mode: Literal["auto", "normal", "low_latency"] = "auto"
-    disable_deepgemm_ue8m0: bool = False
     ep_num_redundant_experts: int = 0
     ep_dispatch_algorithm: Optional[Literal["static", "dynamic", "fake"]] = None
     init_expert_location: str = "trivial"
@@ -1563,11 +1562,6 @@ class ServerArgs:
             default="auto",
             help="Select the mode when enable DeepEP MoE, could be `normal`, `low_latency` or `auto`. Default is `auto`, which means `low_latency` for decode batch and `normal` for prefill batch.",
         )
-        parser.add_argument(
-            "--disable-deepgemm-ue8m0",
-            action="store_true",
-            help="Disable DeepGEMM UE8M0 scaling optimizations. This can help with accuracy issues on Blackwell GPUs (B200) for certain models like DeepSeek.",
-        )
         parser.add_argument(
             "--ep-num-redundant-experts",
             type=int,