[CI] Fix ci tests (#5769)

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

@@ -8,6 +8,7 @@ from typing import Callable, Optional
 import torch
 from torch.nn import functional as F
 
+from sglang.srt.layers.activation import GeluAndMul, SiluAndMul
 from sglang.srt.layers.moe.topk import select_experts
 
 
@@ -30,7 +31,7 @@ def fused_moe_forward_native(
 ) -> torch.Tensor:
 
     if apply_router_weight_on_input:
-        raise NotImplementedError
+        raise NotImplementedError()
 
     topk_weights, topk_ids = select_experts(
         hidden_states=x,
@@ -75,9 +76,6 @@ def moe_forward_native(
     activation: str = "silu",
     routed_scaling_factor: Optional[float] = None,
 ) -> torch.Tensor:
-
-    from sglang.srt.layers.activation import GeluAndMul, SiluAndMul
-
     topk_weights, topk_ids = select_experts(
         hidden_states=x,
         router_logits=router_logits,

python/sglang/srt/layers/moe/fused_moe_triton/fused_moe.py

@@ -13,7 +13,16 @@ import triton
 import triton.language as tl
 
 from sglang.srt.layers.moe.topk import select_experts
-from sglang.srt.layers.quantization.fp8_kernel import scaled_fp8_quant
+from sglang.srt.layers.quantization.fp8_kernel import (
+    per_token_group_quant_fp8,
+    scaled_fp8_quant,
+    sglang_per_token_group_quant_fp8,
+)
+from sglang.srt.layers.quantization.int8_kernel import (
+    per_token_group_quant_int8,
+    per_token_quant_int8,
+    sglang_per_token_group_quant_int8,
+)
 from sglang.srt.utils import (
     direct_register_custom_op,
     get_bool_env_var,
@@ -746,21 +755,6 @@ def invoke_fused_moe_kernel(
     block_shape: Optional[List[int]] = None,
     no_combine: bool = False,
 ) -> None:
-    from sglang.srt.layers.quantization.int8_kernel import (
-        per_token_group_quant_int8,
-        per_token_quant_int8,
-    )
-
-    if _is_cuda:
-        from sglang.srt.layers.quantization.fp8_kernel import (
-            sglang_per_token_group_quant_fp8,
-        )
-        from sglang.srt.layers.quantization.int8_kernel import (
-            sglang_per_token_group_quant_int8,
-        )
-    else:
-        from sglang.srt.layers.quantization.fp8_kernel import per_token_group_quant_fp8
-
     assert topk_weights.stride(1) == 1
     assert sorted_token_ids.stride(0) == 1