[PD][MoRI] Drive KV transfers with a sharded synchronous worker pool (#26922)

This commit is contained in:
Niko Ma
2026-06-08 00:49:25 -07:00
committed by GitHub
parent 0d0254c9de
commit 18d728967a
4 changed files with 235 additions and 112 deletions
@@ -8,7 +8,6 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
PDDisaggregationServerBase,
)
from sglang.test.test_utils import (
DEFAULT_HYBRID_MAMBA_MODEL_NAME_FOR_TEST,
DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
popen_launch_pd_server,
@@ -179,19 +178,5 @@ class TestMoriTransferEngineTPMismatchE2E(MoriTransferEngineBase):
self._assert_generate_smoke()
class TestMoriTransferEngineHybridMambaE2E(MoriTransferEngineBase):
port_delta = 20
prefill_tp = 4
decode_tp = 4
decode_base_gpu_id = 4
required_gpus = 8
model_default = DEFAULT_HYBRID_MAMBA_MODEL_NAME_FOR_TEST
model_env_var = "SGLANG_MORI_HYBRID_E2E_TEST_MODEL"
def test_generate_smoke_hybrid_mamba(self):
self._assert_generate_smoke()
if __name__ == "__main__":
unittest.main()