warn to warning (#6952)

`warn` is deprecated, see
https://docs.python.org/3/library/logging.html#logging.Logger.warning


```DeprecationWarning: The 'warn' method is deprecated, use 'warning' instead```

accelerator/real_accelerator.py

@@ -178,7 +178,7 @@ def get_accelerator():
         if accelerator_name is None:
             # borrow this log from PR#5084
             if accel_logger is not None:
-                accel_logger.warn(
+                accel_logger.warning(
                     "Setting accelerator to CPU. If you have GPU or other accelerator, we were unable to detect it.")
             # cpu added as catch-all when accelerator detection fails
             accelerator_name = "cpu"

deepspeed/runtime/base_optimizer.py

@@ -28,7 +28,7 @@ def load_hp_checkpoint_state_from_checkpoint_dir(self, lp_groups_name: str, chec
 
         tp_rank = bwc_tensor_model_parallel_rank(mpu=self.mpu)
         if self.mpu is None:
-            logger.warn("MPU is not provided, setting tp size to 1 in checkpoint loading.")
+            logger.warning("MPU is not provided, setting tp size to 1 in checkpoint loading.")
             tp_world_size = 1
         else:
             tp_world_size = self.mpu.get_slice_parallel_world_size() if hasattr(self.mpu, "get_slice_parallel_world_size") \

deepspeed/runtime/engine.py

@@ -3120,7 +3120,7 @@ def _get_all_zero_checkpoints(self, load_dir, tag):
                 if bf16_mode is not self.bfloat16_enabled():
                     checkpoint_bit16 = BFLOAT16 if bf16_mode else FP16
                     engine_bit16 = BFLOAT16 if self.bfloat16_enabled() else FP16
-                    logger.warn(f'Loading {checkpoint_bit16} zero checkpoints into {engine_bit16} training engine')
+                    logger.warning(f'Loading {checkpoint_bit16} zero checkpoints into {engine_bit16} training engine')
                 return self._get_all_zero_checkpoint_state_dicts(zero_ckpt_names)
 
         return None
@@ -3276,7 +3276,7 @@ def _save_moe_checkpoint(self, save_dir, tag, client_state={}, exclude_frozen_pa
 
                     local_expert_id = None
                     if not m:
-                        logger.warn(f'No expert found in key {key}.')
+                        logger.warning(f'No expert found in key {key}.')
                     else:
                         local_expert_id = m.group(1)
 

deepspeed/runtime/lr_schedules.py

@@ -508,7 +508,7 @@ def _initialize_lr(self, optimizer, cycle_min_lr, cycle_max_lr, decay_lr_rate, l
     def _initialize_momentum(self, optimizer, cycle_min_mom, cycle_max_mom, decay_mom_rate, last_batch_iteration):
         if 'betas' not in optimizer.defaults:
             optimizer_name = type(optimizer).__name__
-            logger.warn(
+            logger.warning(
                 f"cycle_momentum is disabled because optimizer {optimizer_name} does not support momentum, no betas attribute in defaults"
             )
             self.cycle_momentum = False

deepspeed/runtime/zero/stage_1_and_2.py

@@ -614,7 +614,7 @@ def _configure_moe_settings(self):
             assert self.contiguous_gradients, "Contiguous Gradients in ZeRO Stage 2 must be set to True for MoE. Other code paths are not tested with MoE"
         # NOTE: To run ZeRO stage 1 with MoE, we need to set self.contiguous_gradients to True or ignore the assertion
         if not self.partition_gradients and not self.contiguous_gradients:
-            logger.warn(
+            logger.warning(
                 "ZeRO Stage 1 has not been thoroughly tested with MoE. This configuration is still experimental.")
         assert self.reduce_scatter, "Reduce Scatter in ZeRO Stage 2 must be set to True for MoE. Other code paths are not tested with MoE"