Add singleton logger (#873)

* add singleton wandb logger

* add trainer update

* update comment

* update comment

src/fairchem/core/common/logger.py

@@ -148,3 +148,89 @@ def log_summary(self, summary_dict: dict[str, Any]) -> None:
 
     def log_artifact(self, name: str, type: str, file_location: str) -> None:
         logging.warning("log_artifact for Tensorboard not supported")
+
+
+class WandBSingletonLogger:
+    """
+    Singleton version of wandb logger, this forces a single instance of the logger to be created and used from anywhere in the code (not just from the trainer).
+    This will replace the original WandBLogger.
+
+    We initialize wandb instance somewhere in the trainer/runner globally:
+
+    WandBSingletonLogger.init_wandb(...)
+
+    Then from anywhere in the code we can fetch the singleton instance and log to wandb,
+    note this allows you to log without knowing explicitly which step you are on
+    see: https://docs.wandb.ai/ref/python/log/#the-wb-step for more details
+
+    WandBSingletonLogger.get_instance().log({"some_value": value}, commit=False)
+    """
+
+    _instance = None
+
+    def __init__(self):
+        raise RuntimeError("Call get_instance() instead")
+
+    @classmethod
+    def init_wandb(
+        cls,
+        config: dict,
+        run_id: str,
+        run_name: str,
+        log_dir: str,
+        project: str,
+        entity: str,
+        group: str | None = None,
+    ) -> None:
+        wandb.init(
+            config=config,
+            id=run_id,
+            name=run_name,
+            dir=log_dir,
+            project=project,
+            entity=entity,
+            resume="allow",
+            group=group,
+        )
+
+    @classmethod
+    def get_instance(cls):
+        assert wandb.run is not None, "wandb is not initialized, call init_wandb first!"
+        if cls._instance is None:
+            cls._instance = cls.__new__(cls)
+        return cls._instance
+
+    def watch(self, model, log_freq: int = 1000) -> None:
+        wandb.watch(model, log_freq=log_freq)
+
+    def log(
+        self, update_dict: dict, step: int | None = None, commit=False, split: str = ""
+    ) -> None:
+        # HACK: this is really ugly logic here for backward compat but we should get rid of.
+        # the split string shouldn't inserted here
+        if split != "":
+            new_dict = {}
+            for key in update_dict:
+                new_dict[f"{split}/{key}"] = update_dict[key]
+            update_dict = new_dict
+
+        # if step is not specified, wandb will use an auto-incremented step: https://docs.wandb.ai/ref/python/log/
+        # otherwise the user must increment it manually (not recommended)
+        wandb.log(update_dict, step=step, commit=commit)
+
+    def log_plots(self, plots, caption: str = "") -> None:
+        assert isinstance(plots, list)
+        plots = [wandb.Image(x, caption=caption) for x in plots]
+        wandb.log({"data": plots})
+
+    def log_summary(self, summary_dict: dict[str, Any]):
+        for k, v in summary_dict.items():
+            wandb.run.summary[k] = v
+
+    def mark_preempting(self) -> None:
+        wandb.mark_preempting()
+
+    def log_artifact(self, name: str, type: str, file_location: str) -> None:
+        art = wandb.Artifact(name=name, type=type)
+        art.add_file(file_location)
+        art.save()

src/fairchem/core/trainers/base_trainer.py

@@ -31,6 +31,7 @@
 from fairchem.core import __version__
 from fairchem.core.common import distutils, gp_utils
 from fairchem.core.common.data_parallel import BalancedBatchSampler
+from fairchem.core.common.logger import WandBSingletonLogger
 from fairchem.core.common.registry import registry
 from fairchem.core.common.slurm import (
     add_timestamp_id_to_submission_pickle,
@@ -275,7 +276,19 @@ def load_logger(self) -> None:
             logger_name = logger if isinstance(logger, str) else logger["name"]
             assert logger_name, "Specify logger name"
 
-            self.logger = registry.get_logger_class(logger_name)(self.config)
+            if logger_name == "wandb_singleton":
+                WandBSingletonLogger.init_wandb(
+                    config=self.config,
+                    run_id=self.config["cmd"]["timestamp_id"],
+                    run_name=self.config["cmd"]["identifier"],
+                    log_dir=self.config["cmd"]["logs_dir"],
+                    project=self.config["logger"]["project"],
+                    entity=self.config["logger"]["entity"],
+                    group=self.config["logger"].get("group", ""),
+                )
+                self.logger = WandBSingletonLogger.get_instance()
+            else:
+                self.logger = registry.get_logger_class(logger_name)(self.config)
 
     def get_sampler(
         self, dataset, batch_size: int, shuffle: bool