ci: decouple stage and runner for cuda registry (#25197)

test/registered/quant/test_awq.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=160, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=160, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=200, suite="stage-b-test-1-gpu-large-amd")
 
 

test/registered/quant/test_block_int8.py

@@ -10,7 +10,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=44, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=44, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=22, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/quant/test_deepseek_v32_fp4_mtp_4gpu.py

@@ -17,7 +17,8 @@ from sglang.test.test_utils import (
 
 register_cuda_ci(
     est_time=1060,
-    suite="stage-c-test-4-gpu-b200",
+    stage="stage-c",
+    runner_config="4-gpu-b200",
 )
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3.2-NVFP4"

test/registered/quant/test_deepseek_v3_fp4_4gpu.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
-register_cuda_ci(est_time=1190, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=1190, stage="stage-c", runner_config="4-gpu-b200")
 
 FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
 SERVER_LAUNCH_TIMEOUT = 1200

test/registered/quant/test_fp8_blockwise_gemm.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=630, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=630, stage="stage-c", runner_config="4-gpu-b200")
 
 MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8"
 MXFP8_MODEL_PATH = "zianglih/Qwen3-4B-Instruct-2507-MXFP8"

test/registered/quant/test_fp8_gemm_sm120.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=146, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=146, stage="stage-b", runner_config="1-gpu-small")
 
 PERTENSOR_MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-FP8"
 BLOCKWISE_MODEL_PATH = "Qwen/Qwen3-4B-Instruct-2507-FP8"

test/registered/quant/test_fp8_kernel.py

@@ -9,7 +9,7 @@ from sglang.srt.layers.quantization.fp8_kernel import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=10, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=10, stage="stage-b", runner_config="1-gpu-large")
 
 from sglang.srt.utils import get_device, is_cuda, is_xpu
 

test/registered/quant/test_fp8_utils.py

@@ -10,7 +10,7 @@ from sglang.srt.layers.quantization.fp8_utils import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=9, stage="stage-b", runner_config="1-gpu-large")
 
 
 class TestInverseTransformScaleUe8m0(CustomTestCase):

test/registered/quant/test_fp8kv_triton.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=73, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=73, stage="stage-b", runner_config="1-gpu-large")
 
 
 class TestFP8KVCacheTritonBackend(CustomTestCase):

test/registered/quant/test_gguf.py

@@ -6,7 +6,7 @@ import sglang as sgl
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=76, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=76, stage="stage-b", runner_config="1-gpu-small")
 
 
 class TestGGUF(CustomTestCase):

test/registered/quant/test_gptqmodel_dynamic.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=100, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=100, stage="stage-b", runner_config="1-gpu-large")
 
 
 def check_quant_method(model_path: str, use_marlin_kernel: bool):

test/registered/quant/test_int8_kernel.py

@@ -11,7 +11,7 @@ from sglang.srt.server_args import ServerArgs, set_global_server_args_for_schedu
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=15, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=15, stage="stage-b", runner_config="1-gpu-small")
 
 
 def native_w8a8_per_token_matmul(A, B, As, Bs, output_dtype=torch.float16):

test/registered/quant/test_marlin_moe.py

@@ -12,7 +12,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_marlin_utils import awq_marlin_quantize, marlin_quantize
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=108, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=108, stage="stage-b", runner_config="1-gpu-small")
 
 set_global_server_args_for_scheduler(object.__new__(ServerArgs))
 

test/registered/quant/test_modelopt_fp8.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=62, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=62, stage="stage-b", runner_config="1-gpu-large")
 
 
 class TestModeloptFP8(CustomTestCase):

test/registered/quant/test_nvfp4_gemm.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=420, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=420, stage="stage-c", runner_config="4-gpu-b200")
 
 MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-NVFP4"
 

test/registered/quant/test_nvfp4_gemm_sm120.py

@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
     try_cached_model,
 )
 
-register_cuda_ci(est_time=109, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=109, stage="stage-b", runner_config="1-gpu-small")
 
 MODEL_PATH = "nvidia/Llama-3.1-8B-Instruct-NVFP4"
 

test/registered/quant/test_triton_scaled_mm.py

@@ -9,7 +9,7 @@ from sglang.srt.utils.common import get_device
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_cuda_ci(est_time=11, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=11, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=12, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/quant/test_w8a8_quantization.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=232, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=232, stage="stage-b", runner_config="1-gpu-large")
 
 
 class BaseW8A8Test(CustomTestCase):