[NVIDIA] Fix broken fp8 MoE of deepseek v3 (#13264)

Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
Co-authored-by: Kangyan-Zhou <zky314343421@gmail.com>

python/sglang/srt/layers/quantization/fp8.py

@@ -1210,9 +1210,7 @@ class Fp8MoEMethod(FusedMoEMethodBase):
             else topk_config.correction_bias.to(x.dtype)
         )
 
-        routing_method_type = getattr(
-            layer, "routing_method_type", RoutingMethodType.DeepSeekV3
-        )
+        routing_method_type = getattr(layer, "routing_method_type")
 
         with use_symmetric_memory(
             get_tp_group(), disabled=not is_allocation_symmetric()

python/sglang/srt/models/deepseek_v2.py

@@ -93,6 +93,7 @@ from sglang.srt.layers.moe.ep_moe.layer import DeepEPMoE, get_moe_impl_class
 from sglang.srt.layers.moe.fused_moe_triton.layer import FusedMoE
 from sglang.srt.layers.moe.kt_ep_wrapper import KTEPWrapperMethod
 from sglang.srt.layers.moe.topk import TopK, TopKOutputFormat
+from sglang.srt.layers.moe.utils import RoutingMethodType
 from sglang.srt.layers.quantization.base_config import QuantizationConfig
 from sglang.srt.layers.quantization.fp8 import Fp8Config
 from sglang.srt.layers.quantization.fp8_kernel import (
@@ -674,6 +675,7 @@ class DeepseekV2MoE(nn.Module):
             layer_id=self.layer_id,
             quant_config=quant_config,
             routed_scaling_factor=self.routed_scaling_factor,
+            routing_method_type=RoutingMethodType.DeepSeekV3,
             prefix=add_prefix("experts", prefix),
         )