[CI] improve disaggregation CI. (#11264)

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

View File

@@ -70,11 +70,8 @@ class DisaggregationHiCacheBase(TestDisaggregationBase):
"wait_complete",
"--mem-fraction-static",
"0.8",
"--disaggregation-ib-device",
"mlx5_roce0",
"--disaggregation-transfer-backend",
"mooncake",
]
prefill_args += cls.transfer_backend + cls.rdma_devices
env = {
**os.environ,
"SGLANG_HICACHE_FILE_BACKEND_STORAGE_DIR": cls.temp_dir,
@@ -148,11 +145,8 @@ class TestDisaggregationPrefillWithHiCache(DisaggregationHiCacheBase):
"0.8",
"--base-gpu-id",
"1",
"--disaggregation-ib-device",
"mlx5_roce0",
"--disaggregation-transfer-backend",
"mooncake",
]
decode_args += cls.transfer_backend + cls.rdma_devices
env = {
**os.environ,
"SGLANG_HICACHE_FILE_BACKEND_STORAGE_DIR": cls.temp_dir,
@@ -201,10 +195,6 @@ class TestDisaggregationDecodeWithHiCache(DisaggregationHiCacheBase):
"0.8",
"--base-gpu-id",
"1",
"--disaggregation-ib-device",
"mlx5_roce0",
"--disaggregation-transfer-backend",
"mooncake",
"--disaggregation-decode-enable-offload-kvcache",
"--hicache-ratio",
"1.2",
@@ -215,6 +205,7 @@ class TestDisaggregationDecodeWithHiCache(DisaggregationHiCacheBase):
"--hicache-storage-prefetch-policy",
"wait_complete",
]
decode_args += cls.transfer_backend + cls.rdma_devices
env = {
**os.environ,
"SGLANG_HICACHE_FILE_BACKEND_STORAGE_DIR": cls.temp_dir,

View File

@@ -15,6 +15,7 @@ import requests
from test_hicache_storage_file_backend import HiCacheStorageBaseMixin
from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
from sglang.test.test_disaggregation_utils import get_rdma_devices_args
from sglang.test.test_utils import (
DEFAULT_MLA_MODEL_NAME_FOR_TEST,
CustomTestCase,
@@ -192,7 +193,7 @@ class HiCacheStorageMooncakeBackendBaseMixin(HiCacheStorageBaseMixin):
"""Get additional server arguments specific to configuration - override in subclasses"""
server_args = {
"--tp-size": 1,
"--tp-size": 2,
"--hicache-ratio": 2,
"--hicache-storage-backend": "mooncake",
}
@@ -202,7 +203,7 @@ class HiCacheStorageMooncakeBackendBaseMixin(HiCacheStorageBaseMixin):
"MOONCAKE_MASTER": f"127.0.0.1:{cls.mooncake_master_port}",
"MOONCAKE_PROTOCOL": "rdma",
"MC_MS_AUTO_DISC": "0",
"MOONCAKE_DEVICE": "mlx5_roce0,mlx5_roce1",
"MOONCAKE_DEVICE": get_rdma_devices_args(),
"MOONCAKE_TE_META_DATA_SERVER": f"http://127.0.0.1:{cls.mooncake_metadata_port}/metadata",
"MOONCAKE_GLOBAL_SEGMENT_SIZE": "4294967296", # 4 GiB
}