fix moe gate dtype, fix tbo, fix fake dispatch

python/sglang/srt/eplb/expert_location_dispatch.py

@@ -66,7 +66,7 @@ def transform_select_experts_inputs(
     info: Optional[ExpertLocationDispatchInfo],
 ):
     if (info is not None) and (info.ep_dispatch_algorithm == "fake"):
-        router_logits = torch.randn_like(router_logits)
+        router_logits.uniform_(5, 10)
         if correction_bias is not None:
             correction_bias = torch.zeros_like(correction_bias)
     return router_logits, correction_bias

python/sglang/srt/layers/moe/topk.py

@@ -325,7 +325,7 @@ def biased_grouped_topk_gpu(
         and is_power_of_two(correction_bias.shape[0])
     ):
         topk_weights, topk_ids = moe_fused_gate(
-            gating_output,
+            gating_output.to(dtype=torch.float32),
             correction_bias,
             num_expert_group,
             topk_group,

python/sglang/srt/models/deepseek_v2.py

@@ -224,7 +224,7 @@ def __init__(
         )
         if config.topk_method == "noaux_tc":
             self.e_score_correction_bias = nn.Parameter(
-                torch.empty((config.n_routed_experts))
+                torch.empty((config.n_routed_experts), dtype=torch.float32)
             )
         else:
             self.e_score_correction_bias = None