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

24
test/srt/test_disaggregation_different_tp.py
View File

@@ -41,9 +41,8 @@ class TestDisaggregationMooncakePrefillLargerTP(TestDisaggregationBase):
"prefill",
"--tp",
"4",
"--disaggregation-ib-device",
"mlx5_roce0,mlx5_roce1",
]
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
cls.model,
cls.prefill_url,
@@ -61,9 +60,8 @@ class TestDisaggregationMooncakePrefillLargerTP(TestDisaggregationBase):
"2",
"--base-gpu-id",
"4",
"--disaggregation-ib-device",
"mlx5_roce4,mlx5_roce5",
]
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
cls.model,
cls.decode_url,
@@ -115,9 +113,8 @@ class TestDisaggregationMooncakeDecodeLargerTP(TestDisaggregationBase):
"prefill",
"--tp",
"2",
"--disaggregation-ib-device",
"mlx5_roce0,mlx5_roce1",
]
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
cls.model,
cls.prefill_url,
@@ -135,9 +132,8 @@ class TestDisaggregationMooncakeDecodeLargerTP(TestDisaggregationBase):
"4",
"--base-gpu-id",
"4",
"--disaggregation-ib-device",
"mlx5_roce4,mlx5_roce5",
]
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
cls.model,
cls.decode_url,
@@ -189,9 +185,8 @@ class TestDisaggregationMooncakeMHAPrefillLargerTP(TestDisaggregationBase):
"prefill",
"--tp",
"4",
"--disaggregation-ib-device",
"mlx5_roce0,mlx5_roce1",
]
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
cls.model,
cls.prefill_url,
@@ -209,9 +204,8 @@ class TestDisaggregationMooncakeMHAPrefillLargerTP(TestDisaggregationBase):
"2",
"--base-gpu-id",
"4",
"--disaggregation-ib-device",
"mlx5_roce4,mlx5_roce5",
]
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
cls.model,
cls.decode_url,
@@ -263,9 +257,8 @@ class TestDisaggregationMooncakeMHADecodeLargerTP(TestDisaggregationBase):
"prefill",
"--tp",
"2",
"--disaggregation-ib-device",
"mlx5_roce0,mlx5_roce1",
]
prefill_args += cls.transfer_backend + cls.rdma_devices
cls.process_prefill = popen_launch_pd_server(
cls.model,
cls.prefill_url,
@@ -283,9 +276,8 @@ class TestDisaggregationMooncakeMHADecodeLargerTP(TestDisaggregationBase):
"4",
"--base-gpu-id",
"4",
"--disaggregation-ib-device",
"mlx5_roce4,mlx5_roce5",
]
decode_args += cls.transfer_backend + cls.rdma_devices
cls.process_decode = popen_launch_pd_server(
cls.model,
cls.decode_url,