Pass lint

zeus/policy/optimizer.py

@@ -137,9 +137,7 @@ def observe(
             #       because sampling from an infinite precision Gaussian distribution
             #       always returns the mean (the observation), and it will hamper
             #       exploration in the early stage.
-            precision = (
-                np.inf if variance == 0.0 else np.reciprocal(variance)
-            )  # ruff: noqa: PLR2004
+            precision = np.inf if variance == 0.0 else np.reciprocal(variance)
             mab = self.mabs[job]
             mab.arm_reward_prec[batch_size] = precision
             mab.fit_arm(batch_size, arm_rewards, reset=True)

zeus/run/dataloader.py

@@ -32,7 +32,7 @@
 from torch.utils.data.distributed import DistributedSampler
 
 from zeus.monitor import ZeusMonitor, Measurement
-from zeus.util.check import get_env
+from zeus.util.env import get_env
 from zeus.util.metric import ZeusCostThresholdExceededError, zeus_cost
 from zeus.util.logging import get_logger
 
@@ -957,14 +957,13 @@ def __iter__(self):
             # Push profiling window for the current epoch.
             # Note that both train and eval dataloaders will push one profiling window *separately*.
             self._begin_measurement("__ZeusDataLoader_epoch")
-            # The power limit of the GPU is only changed by the train dataloader.
-            if self._is_train:  # ruff: noqa: SIM102
-                # If we're not profiling, use the steady state power limit.
-                # If we are profiling, the power limit will be set in __next__ with warmup.
-                # Power limit result is already loaded in when initializing the train dataloader,
-                # so we just set the power limit directly.
-                if not self._should_profile:
-                    self._set_gpu_steady_power_limit()
+            # The power limit of the GPU is only changed by the train dataloader (`self._is_train`).
+            # If we're not profiling, use the steady state power limit (`self._should_profile`).
+            # If we are profiling, the power limit will be set in __next__ with warmup.
+            # Power limit result is already loaded in when initializing the train dataloader,
+            # so we just set the power limit directly.
+            if self._is_train and not self._should_profile:
+                self._set_gpu_steady_power_limit()
 
         return self
 

zeus/run/master.py

@@ -271,7 +271,7 @@ def run(
             raise ValueError("You must provide a default batch size for the job.")
         if job.command is None:
             raise ValueError("You must provide a command format string for the job.")
-        if eta_knob < 0.0 or eta_knob > 1.0:  # ruff: noqa: PLR2004
+        if eta_knob < 0.0 or eta_knob > 1.0:
             raise ValueError("eta_knob must be in [0.0, 1.0].")
 
         print(f"[Zeus Master] {job} x {num_recurrence}")