[CI] Reorganize stage-b 1-GPU tests for 5090 compatibility (#16826)

test/registered/spec/eagle/test_eagle3_basic.py

@@ -12,7 +12,6 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=50, suite="stage-b-test-small-1-gpu")
-register_cuda_ci(est_time=87, suite="stage-b-test-small-1-gpu-5090")
 
 
 class TestEagle3Basic(EagleServerBase):

test/registered/spec/eagle/test_eagle_constrained_decoding.py

@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=100, suite="stage-b-test-small-1-gpu")
+register_cuda_ci(est_time=100, suite="stage-b-test-large-1-gpu")
 
 
 class TestEagleConstrainedDecoding(

test/registered/spec/eagle/test_eagle_infer_a.py

@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
-register_cuda_ci(est_time=561, suite="stage-b-test-small-1-gpu")
+register_cuda_ci(est_time=561, suite="stage-b-test-large-1-gpu")
 
 torch_dtype = torch.float16
 prefill_tolerance = 5e-2

test/registered/spec/eagle/test_eagle_infer_b.py

@@ -17,7 +17,7 @@ from sglang.test.kits.radix_cache_server_kit import run_radix_attention_test
 from sglang.test.server_fixtures.eagle_fixture import EagleServerBase
 from sglang.test.test_utils import DEFAULT_TARGET_MODEL_EAGLE, run_logprob_check
 
-register_cuda_ci(est_time=1100, suite="stage-b-test-small-1-gpu")
+register_cuda_ci(est_time=1100, suite="stage-b-test-large-1-gpu")
 
 
 class TestEAGLEServerBasic(EagleServerBase):

test/registered/spec/eagle/test_eagle_infer_beta.py

@@ -17,7 +17,6 @@ from sglang.test.test_utils import (
 )
 
 register_cuda_ci(est_time=283, suite="stage-b-test-small-1-gpu")
-register_cuda_ci(est_time=404, suite="stage-b-test-small-1-gpu-5090")
 
 
 class TestEagleServerBase(CustomTestCase, MatchedStopMixin):