[CI] Reorganize stage-b 1-GPU tests for 5090 compatibility (#16826)
This commit is contained in:
@@ -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):
|
||||
|
||||
@@ -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(
|
||||
|
||||
@@ -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
|
||||
|
||||
@@ -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):
|
||||
|
||||
@@ -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):
|
||||
|
||||
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
|
||||
popen_launch_server,
|
||||
)
|
||||
|
||||
register_cuda_ci(est_time=230, suite="stage-b-test-small-1-gpu")
|
||||
register_cuda_ci(est_time=230, suite="stage-b-test-large-1-gpu")
|
||||
|
||||
GSM_DATASET_PATH = None
|
||||
|
||||
|
||||
@@ -18,7 +18,7 @@ from sglang.test.test_utils import (
|
||||
)
|
||||
|
||||
# Standalone speculative decoding tests (FA3, Triton, FlashInfer backends)
|
||||
register_cuda_ci(est_time=308, suite="stage-b-test-small-1-gpu")
|
||||
register_cuda_ci(est_time=308, suite="stage-b-test-large-1-gpu")
|
||||
|
||||
GSM_DATASET_PATH = None
|
||||
|
||||
|
||||
@@ -10,7 +10,6 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
|
||||
|
||||
register_cuda_ci(est_time=3, suite="stage-b-test-small-1-gpu")
|
||||
register_amd_ci(est_time=3, suite="stage-b-test-small-1-gpu-amd")
|
||||
register_cuda_ci(est_time=2, suite="stage-b-test-small-1-gpu-5090")
|
||||
|
||||
|
||||
class TestBuildEagleTree(unittest.TestCase):
|
||||
|
||||
Reference in New Issue
Block a user