Turn on setting seed and deterministic flags

for_developers/setup_guide.md

@@ -51,18 +51,26 @@ Please see [requirements-lock.txt](requirements-lock.txt). This is what I got af
 
 - Launch detection task ATSS-R50-FPN template
 
-```console
-otx train +recipe=detection/atss_r50_fpn base.data_dir=tests/assets/car_tree_bug model.otx_model.config.bbox_head.num_classes=3 trainer.max_epochs=50 trainer.check_val_every_n_epoch=10 trainer=gpu base.work_dir=outputs/test_work_dir base.output_dir=outputs/test_output_dir
-```
+  ```console
+  otx train +recipe=detection/atss_r50_fpn base.data_dir=tests/assets/car_tree_bug model.otx_model.config.bbox_head.num_classes=3 trainer.max_epochs=50 trainer.check_val_every_n_epoch=10 trainer=gpu base.work_dir=outputs/test_work_dir base.output_dir=outputs/test_output_dir
+  ```
 
 - Change subset names, e.g., "train" -> "train_16" (for training)
 
-```console
-otx train ... data.train_subset.subset_name=<arbitrary-name> data.val_subset.subset_name=<arbitrary-name> data.test_subset.subset_name=<arbitrary-name>
-```
+  ```console
+  otx train ... data.train_subset.subset_name=<arbitrary-name> data.val_subset.subset_name=<arbitrary-name> data.test_subset.subset_name=<arbitrary-name>
+  ```
 
 - Do test with the best validation model checkpoint
 
-```console
-otx train ... test=true
-```
+  ```console
+  otx train ... test=true
+  ```
+
+- Do experiment with deterministic operations and the fixed seed
+
+  ```console
+  otx train ... trainer.deterministic=True seed=<arbitrary-seed>
+  ```
+
+  `trainer.deterministic=True` might affect to the model performance. Please see [this link](https://lightning.ai/docs/pytorch/stable/common/trainer.html#deterministic). Therefore, it is not recommended to turn on this option for the model performance comparison.

pyproject.toml

@@ -269,6 +269,10 @@ max-returns = 10
 "src/otx/**/*.py" = [
     "ERA001",
 ]
+# See https://github.com/openvinotoolkit/training_extensions/actions/runs/7109500350/job/19354528819?pr=2700
+"src/otx/core/config/**/*.py" = [
+    "UP007"
+]
 
 [tool.ruff.pydocstyle]
 convention = "google"

src/otx/config/train.yaml

@@ -11,6 +11,10 @@ defaults:
 train: true
 test: false
 
+# If set it with an integer value, e.g. `seed: 1`,
+# Lightning derives unique seeds across all dataloader workers and processes for torch, numpy and stdlib random number generators.
+seed: null
+
 hydra:
   searchpath:
     - pkg://otx

src/otx/core/config/__init__.py

@@ -2,6 +2,9 @@
 # SPDX-License-Identifier: Apache-2.0
 #
 """Config data type objects."""
+
+from __future__ import annotations
+
 from dataclasses import dataclass
 from typing import Optional
 
@@ -13,18 +16,26 @@
 
 @dataclass
 class TrainConfig:
-    """DTO for training."""
+    """DTO for training.
+
+    Attributes:
+        seed: If set it with an integer value, e.g. `seed=1`,
+            Lightning derives unique seeds across all dataloader workers and processes
+            for torch, numpy and stdlib random number generators.
+    """
 
     base: BaseConfig
     callbacks: dict
     data: DataModuleConfig
     trainer: TrainerConfig
     model: ModelConfig
     logger: dict
-    recipe: Optional[str]  # noqa: FA100
+    recipe: Optional[str]
     train: bool
     test: bool
 
+    seed: Optional[int] = None
+
 
 def register_configs() -> None:
     """Register DTO as default to hydra."""

src/otx/core/engine/train.py

@@ -47,6 +47,8 @@ def train(cfg: TrainConfig) -> tuple[Trainer, dict[str, Any]]:
     :param cfg: A DictConfig configuration composed by Hydra.
     :return: A tuple with Pytorch Lightning Trainer and Python dict of metrics
     """
+    from lightning import seed_everything
+
     from otx.core.data.module import OTXDataModule
     from otx.core.engine.utils.instantiators import (
         instantiate_callbacks,
@@ -55,8 +57,8 @@ def train(cfg: TrainConfig) -> tuple[Trainer, dict[str, Any]]:
     from otx.core.engine.utils.logging_utils import log_hyperparameters
 
     # set seed for random number generators in pytorch, numpy and python.random
-    # if cfg.get("seed"):
-    #     L.seed_everything(cfg.seed, workers=True)
+    if cfg.seed is not None:
+        seed_everything(cfg.seed, workers=True)
 
     log.info(f"Instantiating datamodule <{cfg.data}>")
     datamodule = OTXDataModule(task=cfg.base.task, config=cfg.data)