[AMD][CI] Use non-gated Qwen3-8B for MI35x disaggregation tests (#28678)
This commit is contained in:
@@ -13,7 +13,6 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
|
|||||||
PDDisaggregationServerBase,
|
PDDisaggregationServerBase,
|
||||||
)
|
)
|
||||||
from sglang.test.test_utils import (
|
from sglang.test.test_utils import (
|
||||||
DEFAULT_MODEL_NAME_FOR_TEST,
|
|
||||||
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||||
popen_launch_pd_server,
|
popen_launch_pd_server,
|
||||||
)
|
)
|
||||||
@@ -36,7 +35,7 @@ class TestDisaggregationAccuracy(PDDisaggregationServerBase):
|
|||||||
print("SGLANG_TEST_RDMA_DEVICE is not set! Running without RDMA.")
|
print("SGLANG_TEST_RDMA_DEVICE is not set! Running without RDMA.")
|
||||||
cls.rdma_devices = []
|
cls.rdma_devices = []
|
||||||
|
|
||||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
|
cls.model = "Qwen/Qwen3-8B"
|
||||||
# DEFAULT_MODEL_NAME_FOR_TEST
|
# DEFAULT_MODEL_NAME_FOR_TEST
|
||||||
|
|
||||||
# Non blocking start servers
|
# Non blocking start servers
|
||||||
@@ -235,7 +234,7 @@ class TestDisaggregationMooncakeFailure(PDDisaggregationServerBase):
|
|||||||
# set DISAGGREGATION_TEST_FAILURE_PROB to simulate failure
|
# set DISAGGREGATION_TEST_FAILURE_PROB to simulate failure
|
||||||
os.environ["DISAGGREGATION_TEST_FAILURE_PROB"] = "0.05"
|
os.environ["DISAGGREGATION_TEST_FAILURE_PROB"] = "0.05"
|
||||||
|
|
||||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
|
cls.model = "Qwen/Qwen3-8B"
|
||||||
|
|
||||||
# Non blocking start servers
|
# Non blocking start servers
|
||||||
cls.start_prefill()
|
cls.start_prefill()
|
||||||
@@ -347,7 +346,7 @@ class TestDisaggregationSimulatedRetract(PDDisaggregationServerBase):
|
|||||||
cls.rdma_devices = []
|
cls.rdma_devices = []
|
||||||
|
|
||||||
os.environ["SGLANG_TEST_RETRACT"] = "true"
|
os.environ["SGLANG_TEST_RETRACT"] = "true"
|
||||||
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
|
cls.model = "Qwen/Qwen3-8B"
|
||||||
|
|
||||||
# Non blocking start servers
|
# Non blocking start servers
|
||||||
cls.start_prefill()
|
cls.start_prefill()
|
||||||
|
|||||||
@@ -9,7 +9,6 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
|
|||||||
PDDisaggregationServerBase,
|
PDDisaggregationServerBase,
|
||||||
)
|
)
|
||||||
from sglang.test.test_utils import (
|
from sglang.test.test_utils import (
|
||||||
DEFAULT_MODEL_NAME_FOR_TEST,
|
|
||||||
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
|
||||||
popen_launch_pd_server,
|
popen_launch_pd_server,
|
||||||
try_cached_model,
|
try_cached_model,
|
||||||
@@ -33,7 +32,7 @@ class TestDisaggregationPrefillPPAccuracy(PDDisaggregationServerBase):
|
|||||||
print("SGLANG_TEST_RDMA_DEVICE is not set! Running without RDMA.")
|
print("SGLANG_TEST_RDMA_DEVICE is not set! Running without RDMA.")
|
||||||
cls.rdma_devices = []
|
cls.rdma_devices = []
|
||||||
|
|
||||||
cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST)
|
cls.model = try_cached_model("Qwen/Qwen3-8B")
|
||||||
|
|
||||||
# Non blocking start servers
|
# Non blocking start servers
|
||||||
cls.start_prefill()
|
cls.start_prefill()
|
||||||
@@ -126,7 +125,7 @@ class TestDisaggregationPrefillPPDynamicChunkAccuracy(PDDisaggregationServerBase
|
|||||||
print("SGLANG_TEST_RDMA_DEVICE is not set! Running without RDMA.")
|
print("SGLANG_TEST_RDMA_DEVICE is not set! Running without RDMA.")
|
||||||
cls.rdma_devices = []
|
cls.rdma_devices = []
|
||||||
|
|
||||||
cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST)
|
cls.model = try_cached_model("Qwen/Qwen3-8B")
|
||||||
|
|
||||||
# Non blocking start servers
|
# Non blocking start servers
|
||||||
cls.start_prefill()
|
cls.start_prefill()
|
||||||
@@ -220,7 +219,7 @@ class TestDisaggregationDecodePPAccuracy(PDDisaggregationServerBase):
|
|||||||
print("SGLANG_TEST_RDMA_DEVICE is not set! Running without RDMA.")
|
print("SGLANG_TEST_RDMA_DEVICE is not set! Running without RDMA.")
|
||||||
cls.rdma_devices = []
|
cls.rdma_devices = []
|
||||||
|
|
||||||
cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST)
|
cls.model = try_cached_model("Qwen/Qwen3-8B")
|
||||||
|
|
||||||
# Non blocking start servers
|
# Non blocking start servers
|
||||||
cls.start_prefill()
|
cls.start_prefill()
|
||||||
|
|||||||
Reference in New Issue
Block a user