Move BestCheckpointConfig to utils/checkpoint.py

Differential Revision: D56455020

tests/framework/callbacks/test_base_checkpointer.py

@@ -31,16 +31,14 @@
 from torchtnt.framework.callbacks.base_checkpointer import (
     BaseCheckpointer as BaseCheckpointer,
 )
-from torchtnt.framework.callbacks.checkpointer_types import (
-    BestCheckpointConfig,
-    RestoreOptions,
-)
+from torchtnt.framework.callbacks.checkpointer_types import RestoreOptions
 from torchtnt.framework.callbacks.lambda_callback import Lambda
 from torchtnt.framework.fit import fit
 from torchtnt.framework.state import State
 
 from torchtnt.framework.train import train
 from torchtnt.framework.unit import AppStateMixin, TrainUnit, TTrainData
+from torchtnt.utils.checkpoint import BestCheckpointConfig
 from torchtnt.utils.distributed import get_global_rank, spawn_multi_process
 from torchtnt.utils.env import init_from_env
 from torchtnt.utils.test_utils import skip_if_not_distributed

torchtnt/framework/callbacks/base_checkpointer.py

@@ -16,10 +16,7 @@
 import torch.distributed as dist
 from pyre_extensions import none_throws
 from torchtnt.framework.callback import Callback
-from torchtnt.framework.callbacks.checkpointer_types import (
-    BestCheckpointConfig,
-    RestoreOptions,
-)
+from torchtnt.framework.callbacks.checkpointer_types import RestoreOptions
 from torchtnt.framework.state import EntryPoint, State
 from torchtnt.framework.unit import AppStateMixin, TEvalUnit, TTrainData, TTrainUnit
 from torchtnt.framework.utils import get_timing_context
@@ -28,6 +25,7 @@
     _metadata_exists,
     _sort_by_metric_value,
     _sort_by_recency,
+    BestCheckpointConfig,
     get_best_checkpoint_path,
     get_checkpoint_dirpaths,
     get_latest_checkpoint_path,

torchtnt/framework/callbacks/checkpointer_types.py

@@ -7,7 +7,7 @@
 # pyre-strict
 
 from dataclasses import dataclass
-from typing import Literal, Optional
+from typing import Optional
 
 
 # TODO: eventually support overriding all knobs
@@ -39,17 +39,3 @@ class RestoreOptions:
     restore_eval_progress: bool = True
     restore_optimizers: bool = True
     restore_lr_schedulers: bool = True
-
-
-@dataclass
-class BestCheckpointConfig:
-    """
-    Config for saving the best checkpoints.
-
-    Args:
-        monitored_metric: Metric to monitor for saving best checkpoints. Must be an numerical or tensor attribute on the unit.
-        mode: One of `min` or `max`. The save file is overwritten based the max or min of the monitored metric.
-    """
-
-    monitored_metric: str
-    mode: Literal["min", "max"] = "min"

torchtnt/framework/callbacks/dcp_saver.py

@@ -23,11 +23,7 @@
 )
 
 from torchtnt.framework.callbacks.base_checkpointer import BaseCheckpointer
-from torchtnt.framework.callbacks.checkpointer_types import (
-    BestCheckpointConfig,
-    KnobOptions,
-    RestoreOptions,
-)
+from torchtnt.framework.callbacks.checkpointer_types import KnobOptions, RestoreOptions
 from torchtnt.framework.state import State
 from torchtnt.framework.unit import (
     AppStateMixin,
@@ -37,6 +33,7 @@
     TTrainUnit,
 )
 from torchtnt.framework.utils import get_timing_context
+from torchtnt.utils.checkpoint import BestCheckpointConfig
 from torchtnt.utils.optimizer import init_optim_state
 from torchtnt.utils.rank_zero_log import rank_zero_info, rank_zero_warn
 from torchtnt.utils.stateful import MultiStateful, Stateful

torchtnt/framework/callbacks/torchsnapshot_saver.py

@@ -22,11 +22,7 @@
 )
 
 from torchtnt.framework.callbacks.base_checkpointer import BaseCheckpointer
-from torchtnt.framework.callbacks.checkpointer_types import (
-    BestCheckpointConfig,
-    KnobOptions,
-    RestoreOptions,
-)
+from torchtnt.framework.callbacks.checkpointer_types import KnobOptions, RestoreOptions
 from torchtnt.framework.state import State
 from torchtnt.framework.unit import (
     AppStateMixin,
@@ -36,6 +32,7 @@
     TTrainUnit,
 )
 from torchtnt.framework.utils import get_timing_context
+from torchtnt.utils.checkpoint import BestCheckpointConfig
 from torchtnt.utils.optimizer import init_optim_state
 from torchtnt.utils.rank_zero_log import rank_zero_info, rank_zero_warn
 from torchtnt.utils.stateful import Stateful

torchtnt/utils/__init__.py

@@ -7,6 +7,7 @@
 # pyre-strict
 
 from .checkpoint import (
+    BestCheckpointConfig,
     CheckpointPath,
     get_best_checkpoint_path,
     get_checkpoint_dirpaths,
@@ -160,4 +161,5 @@
     "get_best_checkpoint_path",
     "get_checkpoint_dirpaths",
     "get_latest_checkpoint_path",
+    "BestCheckpointConfig",
 ]

torchtnt/utils/checkpoint.py

@@ -31,6 +31,20 @@ class MetricData:
     value: float
 
 
+@dataclass
+class BestCheckpointConfig:
+    """
+    Config for saving the best checkpoints.
+
+    Args:
+        monitored_metric: Metric to monitor for saving best checkpoints. Must be an numerical or tensor attribute on the unit.
+        mode: One of `min` or `max`. The save file is overwritten based the max or min of the monitored metric.
+    """
+
+    monitored_metric: str
+    mode: Literal["min", "max"] = "min"
+
+
 @total_ordering
 class CheckpointPath:
     """