Update seed in CPU UTs to avoid flaky failure with single test (#7544)

test/srt/cpu/test_activation.py

@@ -8,7 +8,7 @@ from utils import SiluAndMul, precision
 
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 class TestActivation(CustomTestCase):

test/srt/cpu/test_decode.py

@@ -6,7 +6,7 @@ from torch.nn.functional import scaled_dot_product_attention
 
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 class TestDecodeAttention(CustomTestCase):

test/srt/cpu/test_extend.py

@@ -6,7 +6,7 @@ from torch.nn.functional import scaled_dot_product_attention
 
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 class TestExtendAttention(CustomTestCase):

test/srt/cpu/test_gemm.py

@@ -14,7 +14,7 @@ from utils import (
 
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 class Mod(nn.Module):

test/srt/cpu/test_mla.py

@@ -8,7 +8,7 @@ from utils import precision
 
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 class TestMLA(CustomTestCase):

test/srt/cpu/test_moe.py

@@ -8,7 +8,7 @@ import torch
 
 kernel = torch.ops.sgl_kernel
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 from utils import (
     BLOCK_K,

test/srt/cpu/test_norm.py

@@ -8,7 +8,7 @@ from utils import make_non_contiguous, precision
 
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 class TestNorm(CustomTestCase):

test/srt/cpu/test_qkv_proj_with_rope.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import CustomTestCase
 convert_weight_packed = torch.ops.sgl_kernel.convert_weight_packed
 qkv_proj_with_rope = torch.ops.sgl_kernel.qkv_proj_with_rope
 qkv_proj_with_rope_fused_weight = torch.ops.sgl_kernel.qkv_proj_with_rope_fused_weight
-torch.manual_seed(0)
+torch.manual_seed(1234)
 # constants
 kv_lora_rank = 512
 qk_head_dim = 192

test/srt/cpu/test_rope.py

@@ -10,7 +10,7 @@ from sglang.srt.layers.rotary_embedding import (
 )
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 class TestROPE(CustomTestCase):

test/srt/cpu/test_shared_expert.py

@@ -22,7 +22,7 @@ from utils import (
 
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 class TestSharedExpert(CustomTestCase):

test/srt/cpu/test_topk.py

@@ -13,7 +13,7 @@ from sglang.srt.layers.moe.topk import grouped_topk_gpu as native_grouped_topk
 from sglang.srt.models.llama4 import Llama4MoE
 from sglang.test.test_utils import CustomTestCase
 
-torch.manual_seed(0)
+torch.manual_seed(1234)
 
 
 # This is used by the Deepseek-V2 model