sched

Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>

tests/plugins_tests/test_scheduler_plugins.py

@@ -6,7 +6,7 @@ from vllm.core.scheduler import Scheduler
 from vllm.engine.arg_utils import EngineArgs
 from vllm.engine.llm_engine import LLMEngine
 from vllm.sampling_params import SamplingParams
-from vllm.v1.core.scheduler import Scheduler as V1Scheduler
+from vllm.v1.core.sched.scheduler import Scheduler as V1Scheduler
 from vllm.v1.engine.llm_engine import LLMEngine as V1LLMEngine
 
 

tests/v1/core/test_scheduler.py

@@ -6,7 +6,8 @@ import pytest
 from vllm.config import CacheConfig, ModelConfig, SchedulerConfig, VllmConfig
 from vllm.multimodal.inputs import MultiModalKwargs, PlaceholderRange
 from vllm.sampling_params import SamplingParams
-from vllm.v1.core.scheduler import Scheduler, SchedulerOutput
+from vllm.v1.core.sched.output import SchedulerOutput
+from vllm.v1.core.sched.scheduler import Scheduler
 from vllm.v1.outputs import ModelRunnerOutput
 from vllm.v1.request import Request, RequestStatus
 from vllm.v1.structured_output import StructuredOutputManager

tests/v1/worker/test_gpu_model_runner.py

@@ -3,8 +3,8 @@ import pytest
 
 from vllm.config import CacheConfig, ModelConfig, SchedulerConfig, VllmConfig
 from vllm.sampling_params import SamplingParams
-from vllm.v1.core.scheduler.output import (CachedRequestData, NewRequestData,
-                                           SchedulerOutput)
+from vllm.v1.core.sched.output import (CachedRequestData, NewRequestData,
+                                       SchedulerOutput)
 from vllm.v1.sample.metadata import SamplingMetadata
 from vllm.v1.worker.gpu_model_runner import GPUModelRunner
 

vllm/engine/arg_utils.py

@@ -1440,7 +1440,7 @@ class EngineArgs:
         # V1 should use the new scheduler by default.
         # Swap it only if this arg is set to the original V0 default
         if self.scheduler_cls == EngineArgs.scheduler_cls:
-            self.scheduler_cls = "vllm.v1.core.scheduler.Scheduler"
+            self.scheduler_cls = "vllm.v1.core.sched.scheduler.Scheduler"
 
         # When no user override, set the default values based on the usage
         # context.

vllm/executor/ray_utils.py

@@ -17,7 +17,7 @@ from vllm.utils import get_ip
 from vllm.worker.worker_base import WorkerWrapperBase
 
 if TYPE_CHECKING:
-    from vllm.v1.core.scheduler import SchedulerOutput
+    from vllm.v1.core.sched.output import SchedulerOutput
     from vllm.v1.outputs import ModelRunnerOutput
 
 logger = init_logger(__name__)

vllm/v1/attention/backends/flash_attn.py

@@ -16,7 +16,7 @@ from vllm.platforms import current_platform
 from vllm.utils import cdiv
 
 if TYPE_CHECKING:
-    from vllm.v1.core.scheduler.output import SchedulerOutput
+    from vllm.v1.core.sched.output import SchedulerOutput
     from vllm.v1.worker.gpu_input_batch import InputBatch
     from vllm.v1.worker.gpu_model_runner import GPUModelRunner
 

vllm/v1/attention/backends/mla/common.py

@@ -233,7 +233,7 @@ except ImportError:
     from flash_attn import flash_attn_varlen_func
 
 if TYPE_CHECKING:
-    from vllm.v1.core.scheduler.output import SchedulerOutput
+    from vllm.v1.core.sched.output import SchedulerOutput
     from vllm.v1.worker.gpu_input_batch import InputBatch
     from vllm.v1.worker.gpu_model_runner import GPUModelRunner
 

vllm/v1/core/sched/scheduler.py

@@ -13,8 +13,8 @@ from vllm.logger import init_logger
 from vllm.v1.core.encoder_cache_manager import (EncoderCacheManager,
                                                 compute_encoder_budget)
 from vllm.v1.core.kv_cache_manager import KVCacheManager
-from vllm.v1.core.scheduler.output import (CachedRequestData, NewRequestData,
-                                           SchedulerOutput)
+from vllm.v1.core.sched.output import (CachedRequestData, NewRequestData,
+                                       SchedulerOutput)
 from vllm.v1.engine import (EngineCoreEvent, EngineCoreEventType,
                             EngineCoreOutput, EngineCoreOutputs)
 from vllm.v1.metrics.stats import SchedulerStats

vllm/v1/engine/core.py

@@ -22,7 +22,7 @@ from vllm.transformers_utils.config import (
 from vllm.utils import (get_exception_traceback, resolve_obj_by_qualname,
                         zmq_socket_ctx)
 from vllm.v1.core.kv_cache_utils import get_kv_cache_configs
-from vllm.v1.core.scheduler.output import SchedulerOutput
+from vllm.v1.core.sched.output import SchedulerOutput
 from vllm.v1.engine import (EngineCoreOutputs, EngineCoreRequest,
                             EngineCoreRequestType, UtilityOutput)
 from vllm.v1.engine.mm_input_cache import MMInputCacheServer

vllm/v1/worker/gpu_model_runner.py

@@ -44,7 +44,7 @@ from vllm.v1.worker.lora_model_runner_mixin import LoRAModelRunnerMixin
 if TYPE_CHECKING:
     import xgrammar as xgr
 
-    from vllm.v1.core.scheduler.output import SchedulerOutput
+    from vllm.v1.core.sched.output import SchedulerOutput
 else:
     xgr = LazyLoader("xgr", globals(), "xgrammar")
 

vllm/v1/worker/gpu_worker.py

@@ -28,7 +28,7 @@ from vllm.v1.worker.worker_base import WorkerBase
 logger = init_logger(__name__)
 
 if TYPE_CHECKING:
-    from vllm.v1.core.scheduler.output import SchedulerOutput
+    from vllm.v1.core.sched.output import SchedulerOutput
 
 
 class Worker(WorkerBase):

vllm/v1/worker/tpu_model_runner.py

@@ -35,7 +35,7 @@ from vllm.v1.utils import bind_kv_cache
 from vllm.v1.worker.gpu_input_batch import CachedRequestState, InputBatch
 
 if TYPE_CHECKING:
-    from vllm.v1.core.scheduler.output import SchedulerOutput
+    from vllm.v1.core.sched.output import SchedulerOutput
 
 logger = init_logger(__name__)
 

vllm/v1/worker/tpu_worker.py

@@ -17,7 +17,7 @@ from vllm.distributed import (ensure_model_parallel_initialized,
 from vllm.logger import init_logger
 from vllm.model_executor import set_random_seed
 from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE
-from vllm.v1.core.scheduler.output import SchedulerOutput
+from vllm.v1.core.sched.output import SchedulerOutput
 from vllm.v1.kv_cache_interface import (FullAttentionSpec, KVCacheConfig,
                                         KVCacheSpec)
 from vllm.v1.outputs import ModelRunnerOutput