diff --git a/python/sglang/srt/managers/schedule_batch.py b/python/sglang/srt/managers/schedule_batch.py index de2080049404..15995905ac3f 100644 --- a/python/sglang/srt/managers/schedule_batch.py +++ b/python/sglang/srt/managers/schedule_batch.py @@ -96,8 +96,8 @@ from sglang.srt.configs.model_config import ModelConfig from sglang.srt.managers.hisparse_coordinator import HiSparseCoordinator - from sglang.srt.managers.session_controller import Session from sglang.srt.observability.scheduler_metrics_mixin import PrefillStats + from sglang.srt.session.session_controller import Session from sglang.srt.speculative.eagle_info import EagleDraftInput from sglang.srt.speculative.spec_info import SpecInput, SpeculativeAlgorithm diff --git a/python/sglang/srt/managers/scheduler.py b/python/sglang/srt/managers/scheduler.py index 4d0c6a99c8f3..8c92c92e07ca 100644 --- a/python/sglang/srt/managers/scheduler.py +++ b/python/sglang/srt/managers/scheduler.py @@ -183,12 +183,10 @@ from sglang.srt.managers.scheduler_update_weights_mixin import ( SchedulerUpdateWeightsMixin, ) -from sglang.srt.managers.session_controller import SessionController from sglang.srt.managers.utils import GenerationBatchResult, validate_input_length from sglang.srt.mem_cache.cache_init_params import CacheInitParams from sglang.srt.mem_cache.common import release_kv_cache from sglang.srt.mem_cache.radix_cache import RadixCache -from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache from sglang.srt.model_executor.forward_batch_info import ForwardMode, PPProxyTensors from sglang.srt.model_loader.utils import get_resolved_model_impl from sglang.srt.multiplex.multiplexing_mixin import SchedulerMultiplexMixin @@ -206,6 +204,8 @@ from sglang.srt.parser.reasoning_parser import ReasoningParser from sglang.srt.sampling.sampling_batch_info import SamplingBatchInfo from sglang.srt.server_args import PortArgs, ServerArgs, get_global_server_args +from sglang.srt.session.session_aware_cache import SessionAwareCache +from sglang.srt.session.session_controller import SessionController from sglang.srt.speculative.spec_info import SpeculativeAlgorithm from sglang.srt.utils import ( DynamicGradMode, diff --git a/python/sglang/srt/managers/scheduler_runtime_checker_mixin.py b/python/sglang/srt/managers/scheduler_runtime_checker_mixin.py index 34e7634fce6c..838670356ce2 100644 --- a/python/sglang/srt/managers/scheduler_runtime_checker_mixin.py +++ b/python/sglang/srt/managers/scheduler_runtime_checker_mixin.py @@ -8,8 +8,8 @@ from sglang.srt.disaggregation.utils import DisaggregationMode from sglang.srt.environ import envs -from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache from sglang.srt.observability.metrics_collector import QueueCount +from sglang.srt.session.session_aware_cache import SessionAwareCache from sglang.srt.utils.common import ceil_align, raise_error_or_warn from sglang.srt.utils.request_logger import disable_request_logging from sglang.srt.utils.watchdog import WatchdogRaw diff --git a/python/sglang/srt/session/__init__.py b/python/sglang/srt/session/__init__.py new file mode 100644 index 000000000000..e69de29bb2d1 diff --git a/python/sglang/srt/mem_cache/session_aware_cache.py b/python/sglang/srt/session/session_aware_cache.py similarity index 100% rename from python/sglang/srt/mem_cache/session_aware_cache.py rename to python/sglang/srt/session/session_aware_cache.py diff --git a/python/sglang/srt/managers/session_controller.py b/python/sglang/srt/session/session_controller.py similarity index 99% rename from python/sglang/srt/managers/session_controller.py rename to python/sglang/srt/session/session_controller.py index 889e2d60b3f7..00776ee31a18 100644 --- a/python/sglang/srt/managers/session_controller.py +++ b/python/sglang/srt/session/session_controller.py @@ -24,7 +24,7 @@ TokenizedGenerateReqInput, ) from sglang.srt.managers.schedule_batch import FINISH_ABORT, Req -from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache +from sglang.srt.session.session_aware_cache import SessionAwareCache from sglang.srt.utils.common import log_info_on_rank0 if TYPE_CHECKING: diff --git a/test/registered/unit/mem_cache/test_streaming_session_unit.py b/test/registered/unit/mem_cache/test_streaming_session_unit.py index 0a701dcb5dc7..4a349f9f6e6b 100644 --- a/test/registered/unit/mem_cache/test_streaming_session_unit.py +++ b/test/registered/unit/mem_cache/test_streaming_session_unit.py @@ -5,7 +5,7 @@ from sglang.srt.managers.schedule_batch import FINISH_ABORT from sglang.srt.mem_cache.base_prefix_cache import MatchResult from sglang.srt.mem_cache.common import release_kv_cache -from sglang.srt.mem_cache.session_aware_cache import SessionAwareCache, SessionSlot +from sglang.srt.session.session_aware_cache import SessionAwareCache, SessionSlot from sglang.test.ci.ci_register import register_cpu_ci register_cpu_ci(est_time=8, suite="stage-a-test-cpu")