fix pylint

llumnix/backends/vllm/scheduler.py

@@ -48,8 +48,7 @@ def output_len(self) -> int:
     def inference_type(self) -> bool:
         if self.is_prefill():
             return RequestInferenceType.PREFILL
-        else:
-            return RequestInferenceType.DECODE
+        return RequestInferenceType.DECODE
 
 # TODO(ZeldaHuang): adapt prefix cache and sliding window, now use v1 manager
 class BlockManagerLlumnix(BlockSpaceManagerV1):

llumnix/llumlet/local_migration_scheduler.py

@@ -11,8 +11,9 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import numpy as np
 from typing import List, Optional
+import numpy as np
+
 from llumnix.llumlet.request import LlumnixRequest, RequestInferenceType
 from llumnix.backends.backend_interface import BackendInterface
 
@@ -35,9 +36,9 @@ def get_migrate_out_request(self, min_request_len=0, max_request_len=np.inf) ->
     def get_last_running_request(self, min_request_len, max_request_len):
         running: List[LlumnixRequest] = self.backend_engine.get_running_queue()
         for request in reversed(running):
-                    if request.inference_type == RequestInferenceType.DECODE \
-                        and min_request_len <= request.request_len <= max_request_len:
-                        return request
+            if request.inference_type == RequestInferenceType.DECODE \
+                and min_request_len <= request.request_len <= max_request_len:
+                return request
         return None
     def get_longest_running_request(self, min_request_len, max_request_len):
         running: List[LlumnixRequest] = self.backend_engine.get_running_queue()
@@ -55,4 +56,4 @@ def get_shortest_running_request(self, min_request_len, max_request_len):
 
         shortest_seq_group = min((request for request in running if condition(request)), \
                                 key=lambda request: request.request_len, default=None)
-        return shortest_seq_group
+        return shortest_seq_group

llumnix/llumlet/migration_coordinator.py

@@ -93,10 +93,10 @@ def migrate_out_multistage(self, migrate_in_ray_actor: "ray.actor.ActorHandle",
         """
         state_count = 0
         while state_count < self.max_stages:
-             state_count += 1
-             status = self.migrate_out_onestage(migrate_in_ray_actor, migrate_out_request)
-             if MigrationStatus.is_finished(status):
-                 return status
+            state_count += 1
+            status = self.migrate_out_onestage(migrate_in_ray_actor, migrate_out_request)
+            if MigrationStatus.is_finished(status):
+                return status
         # exceed max stages
         return MigrationStatus.FINISHED_ABORTED
 

llumnix/llumlet/request.py

@@ -52,4 +52,4 @@ def output_len(self) -> int:
     def should_abort_migration(self) -> bool:
         return self.output_len == 0 \
             or (self.last_preemption_time and self.last_preemption_time > self.stage_timestamps[-1]) \
-            or self.inference_type == RequestInferenceType.PREFILL
+            or self.inference_type == RequestInferenceType.PREFILL