remove unecessary dual stream token threshold from the rest of models (qwen moe, kimi linear, etc.) (#14337)

python/sglang/srt/models/bailing_moe.py

@@ -349,11 +349,9 @@ class BailingMoESparseMoeBlock(nn.Module):
         num_tokens, hidden_size = hidden_states.shape
         hidden_states = hidden_states.view(-1, hidden_size)
 
-        DUAL_STREAM_TOKEN_THRESHOLD = 1024
         if (
             self.alt_stream is not None
             and hidden_states.shape[0] > 0
-            and hidden_states.shape[0] <= DUAL_STREAM_TOKEN_THRESHOLD
             and get_is_capture_mode()
         ):
             final_hidden_states, shared_output = self.forward_normal_dual_stream(

python/sglang/srt/models/kimi_linear.py

@@ -125,13 +125,11 @@ class KimiMoE(nn.Module):
         hidden_states = hidden_states.view(-1, hidden_size)
 
         shared_output = None
-        DUAL_STREAM_TOKEN_THRESHOLD = 1024
 
         if (
             self.alt_stream is not None
             and self.num_shared_experts is not None
             and hidden_states.shape[0] > 0
-            and hidden_states.shape[0] <= DUAL_STREAM_TOKEN_THRESHOLD
             and get_is_capture_mode()
         ):
             current_stream = torch.cuda.current_stream()

python/sglang/srt/models/llada2.py

@@ -349,11 +349,9 @@ class LLaDA2MoeSparseMoeBlock(nn.Module):
         num_tokens, hidden_size = hidden_states.shape
         hidden_states = hidden_states.view(-1, hidden_size)
 
-        DUAL_STREAM_TOKEN_THRESHOLD = 1024
         if (
             self.alt_stream is not None
             and hidden_states.shape[0] > 0
-            and hidden_states.shape[0] <= DUAL_STREAM_TOKEN_THRESHOLD
             and get_is_capture_mode()
         ):
             final_hidden_states, shared_output = self.forward_normal_dual_stream(

python/sglang/srt/models/qwen2_moe.py

@@ -304,11 +304,9 @@ class Qwen2MoeSparseMoeBlock(nn.Module):
         if get_moe_a2a_backend().is_deepep():
             return self._forward_deepep(hidden_states, forward_batch)
 
-        DUAL_STREAM_TOKEN_THRESHOLD = 1024
         if (
             self.alt_stream is not None
             and hidden_states.shape[0] > 0
-            and hidden_states.shape[0] <= DUAL_STREAM_TOKEN_THRESHOLD
             and get_is_capture_mode()
         ):
             final_hidden_states, shared_output = self.forward_normal_dual_stream(