EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[CI] improve disaggregation CI. (#11264)

Browse Source 
Signed-off-by: Shangming Cai <csmthu@gmail.com>
Co-authored-by: Shangming Cai <csmthu@gmail.com>
This commit is contained in:
Liangsheng Yin
2025-10-08 21:40:56 +08:00
committed by GitHub
parent e3bb7f5ae6
commit c882b5ae75
9 changed files with 107 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
test/srt/test_disaggregation.py
View File

@@ -40,10 +40,9 @@ class TestDisaggregationAccuracy(TestDisaggregationBase):
"--disaggregation-mode",
"prefill",
"--tp",
"2",
"--disaggregation-ib-device",
"mlx5_roce0,mlx5_roce1",
"1",
]
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
cls.model,
cls.prefill_url,
@@ -58,12 +57,11 @@ class TestDisaggregationAccuracy(TestDisaggregationBase):
"--disaggregation-mode",
"decode",
"--tp",
"2",
"1",
"--base-gpu-id",
"2",
"--disaggregation-ib-device",
"mlx5_roce2,mlx5_roce3",
"1",
]
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
cls.model,
cls.decode_url,
@@ -171,10 +169,9 @@ class TestDisaggregationMooncakeFailure(TestDisaggregationBase):
"--disaggregation-mode",
"prefill",
"--tp",
"2",
"--disaggregation-ib-device",
"mlx5_roce0,mlx5_roce1",
"1",
]
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
cls.model,
cls.prefill_url,
@@ -189,12 +186,11 @@ class TestDisaggregationMooncakeFailure(TestDisaggregationBase):
"--disaggregation-mode",
"decode",
"--tp",
"2",
"1",
"--base-gpu-id",
"2",
"--disaggregation-ib-device",
"mlx5_roce2,mlx5_roce3",
"1",
]
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
cls.model,
cls.decode_url,
@@ -270,10 +266,9 @@ class TestDisaggregationMooncakeSpec(TestDisaggregationBase):
"--disaggregation-mode",
"prefill",
"--tp",
"2",
"--disaggregation-ib-device",
"mlx5_roce0,mlx5_roce1",
"1",
] + cls.spec_args
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
cls.model,
cls.prefill_url,
@@ -288,12 +283,11 @@ class TestDisaggregationMooncakeSpec(TestDisaggregationBase):
"--disaggregation-mode",
"decode",
"--tp",
"2",
"1",
"--base-gpu-id",
"2",
"--disaggregation-ib-device",
"mlx5_roce2,mlx5_roce3",
"1",
] + cls.spec_args
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
cls.model,
cls.decode_url,
@@ -346,10 +340,9 @@ class TestDisaggregationSimulatedRetract(TestDisaggregationBase):
"--disaggregation-mode",
"prefill",
"--tp",
"2",
"--disaggregation-ib-device",
"mlx5_roce0,mlx5_roce1",
"1",
]
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
cls.model,
cls.prefill_url,
@@ -364,12 +357,11 @@ class TestDisaggregationSimulatedRetract(TestDisaggregationBase):
"--disaggregation-mode",
"decode",
"--tp",
"2",
"1",
"--base-gpu-id",
"2",
"--disaggregation-ib-device",
"mlx5_roce2,mlx5_roce3",
"1",
]
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
cls.model,
cls.decode_url,