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

test/registered/rl/test_fp32_lm_head.py

@@ -15,7 +15,7 @@ from sglang.srt.server_args import (
 from sglang.srt.utils import get_device
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=9, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=9, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=15, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/rl/test_lora_load_from_tensor.py

@@ -10,7 +10,7 @@ import sglang as sgl
 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=102, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=102, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=90, suite="stage-b-test-1-gpu-small-amd")
 
 MODEL_PATH = "Qwen/Qwen3-0.6B"

test/registered/rl/test_multi_instance_release_memory_occupation.py

@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
     find_available_port,
 )
 
-register_cuda_ci(est_time=57, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=57, stage="stage-c", runner_config="4-gpu-h100")
 register_amd_ci(
     est_time=64,
     suite="stage-c-test-4-gpu-amd",

test/registered/rl/test_patch_torch.py

@@ -9,7 +9,7 @@ import torch.multiprocessing as mp
 from sglang.srt.utils.patch_torch import monkey_patch_torch_reductions
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=15, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=15, stage="stage-b", runner_config="2-gpu-large")
 
 
 class TestReleaseMemoryOccupation(unittest.TestCase):

test/registered/rl/test_release_memory_occupation.py

@@ -55,7 +55,8 @@ from sglang.test.test_utils import (
 
 register_cuda_ci(
     est_time=200,
-    suite="stage-c-test-4-gpu-h100",
+    stage="stage-c",
+    runner_config="4-gpu-h100",
     disabled="Temporarily disabled - needs investigation",
 )
 

test/registered/rl/test_return_routed_experts.py

@@ -24,7 +24,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=400, suite="stage-c-test-4-gpu-h100")
+register_cuda_ci(est_time=400, stage="stage-c", runner_config="4-gpu-h100")
 
 # FP8 variant of Qwen3-30B-A3B: required because DeepEP normal/LL fast paths in
 # ep_moe/layer.py only run for {Fp8Config (via deep_gemm), W4AFp8Config, aiter,

test/registered/rl/test_update_weights_from_disk.py

@@ -21,7 +21,9 @@ from sglang.test.test_utils import (
 register_amd_ci(
     est_time=210, suite="stage-b-test-1-gpu-small-amd", disabled="see #14021"
 )
-register_cuda_ci(est_time=210, suite="stage-b-test-1-gpu-large", disabled="see #14021")
+register_cuda_ci(
+    est_time=210, stage="stage-b", runner_config="1-gpu-large", disabled="see #14021"
+)
 
 
 ###############################################################################

test/registered/rl/test_update_weights_from_disk_blackwell.py

@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(est_time=400, suite="stage-c-test-4-gpu-b200")
+register_cuda_ci(est_time=400, stage="stage-c", runner_config="4-gpu-b200")
 
 import unittest
 

test/registered/rl/test_update_weights_from_distributed.py

@@ -43,7 +43,7 @@ from sglang.test.test_utils import (
 )
 from sglang.utils import terminate_process
 
-register_cuda_ci(est_time=137, suite="stage-b-test-2-gpu-large")
+register_cuda_ci(est_time=137, stage="stage-b", runner_config="2-gpu-large")
 register_amd_ci(est_time=400, suite="stage-b-test-2-gpu-large-amd")
 
 mp.set_start_method("spawn", force=True)

test/registered/rl/test_update_weights_from_tensor.py

@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
-register_cuda_ci(est_time=147, suite="stage-b-test-1-gpu-small")
+register_cuda_ci(est_time=147, stage="stage-b", runner_config="1-gpu-small")
 register_amd_ci(est_time=195, suite="stage-b-test-1-gpu-small-amd")
 
 import gc