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

test/registered/mla/test_flashmla.py

@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
 )
 
 # FlashMLA attention backend tests with MTP speculative decoding
-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")
 
 
 class TestFlashMLAMTP(CustomTestCase):

test/registered/mla/test_mla_flashinfer.py

@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
 )
 
 # FlashInfer MLA backend tests with MTP speculative decoding
-register_cuda_ci(est_time=130, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=130, stage="stage-b", runner_config="1-gpu-large")
 
 
 class TestFlashinferMLAMTP(CustomTestCase):

test/registered/mla/test_mla_fp8.py

@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
 )
 
 # MLA FP8 KV cache test with MGSM evaluation
-register_cuda_ci(est_time=104, suite="stage-b-test-1-gpu-large")
+register_cuda_ci(est_time=104, stage="stage-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=800, suite="stage-b-test-1-gpu-small-amd")
 
 

test/registered/mla/test_mla_int8_deepseek_v3.py

@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
 )
 
 # DeepSeek-V3 INT8 quantization tests (channel and block INT8)
-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")
 
 
 class TestDeepseekV3MTPChannelInt8(CustomTestCase):