From d042ffa3a92c991e9266c294c7b6bc72e9cc27ca Mon Sep 17 00:00:00 2001
From: lbluque <lbluque@meta.com>
Date: Fri, 31 Jan 2025 09:36:53 -0800
Subject: [PATCH 1/2] add git hash changes announcement

---
 src/fairchem/core/_cli_hydra.py        | 36 +++++++++++++++++---------
 src/fairchem/core/components/runner.py |  2 +-
 2 files changed, 25 insertions(+), 13 deletions(-)

diff --git a/src/fairchem/core/_cli_hydra.py b/src/fairchem/core/_cli_hydra.py
index cb08714829..61e92ad80e 100644
--- a/src/fairchem/core/_cli_hydra.py
+++ b/src/fairchem/core/_cli_hydra.py
@@ -50,6 +50,7 @@ class SchedulerConfig:
     mode: SchedulerType = SchedulerType.LOCAL
     ranks_per_node: int = 1
     num_nodes: int = 1
+    num_jobs: int = 1
     slurm: dict = field(
         default_factory=lambda: {
             "mem_gb": 80,  # slurm mem in GB
@@ -83,7 +84,7 @@ def checkpoint_dir(self) -> str:
 
 
 class Submitit(Checkpointable):
-    def __call__(self, dict_config: DictConfig) -> None:
+    def __call__(self, dict_config: DictConfig, **run_kwargs) -> None:
         self.config = dict_config
         job_config: JobConfig = OmegaConf.to_object(dict_config.job)
         # TODO: setup_imports is not needed if we stop instantiating models with Registry.
@@ -94,7 +95,7 @@ def __call__(self, dict_config: DictConfig) -> None:
         runner: Runner = hydra.utils.instantiate(dict_config.runner)
         runner.job_config = job_config
         runner.load_state()
-        runner.run()
+        runner.run(**run_kwargs)
         distutils.cleanup()
 
     def _init_logger(self) -> None:
@@ -151,10 +152,6 @@ def get_hydra_config_from_yaml(
     return hydra.compose(config_name=config_name, overrides=overrides_args)
 
 
-def runner_wrapper(config: DictConfig):
-    Submitit()(config)
-
-
 def main(
     args: argparse.Namespace | None = None, override_args: list[str] | None = None
 ):
@@ -188,10 +185,25 @@ def main(
             slurm_qos=scheduler_cfg.slurm.qos,
             slurm_account=scheduler_cfg.slurm.account,
         )
-        job = executor.submit(runner_wrapper, cfg)
-        logging.info(
-            f"Submitted job id: {job_cfg.timestamp_id}, slurm id: {job.job_id}, logs: {job_cfg.log_dir}"
-        )
+        if scheduler_cfg.num_jobs == 1:
+            job = executor.submit(Submitit(), cfg)
+            logging.info(
+                f"Submitted job id: {job_cfg.timestamp_id}, slurm id: {job.job_id}, logs: {job_cfg.log_dir}"
+            )
+        elif scheduler_cfg.num_jobs > 1:
+            executor.update_parameters(slurm_array_parallelism=scheduler_cfg.num_jobs)
+
+            jobs = []
+            with executor.batch():
+                for job_number in range(scheduler_cfg.num_jobs):
+                    job = executor.submit(
+                        Submitit(),
+                        cfg,
+                        job_number=job_number,
+                        num_jobs=scheduler_cfg.num_jobs,
+                    )
+                    jobs.append(job)
+            logging.info(f"Submitted {len(jobs)} jobs: {jobs[0].job_id.split("_")[0]}")
     else:
         from torch.distributed.launcher.api import LaunchConfig, elastic_launch
 
@@ -204,8 +216,8 @@ def main(
                 rdzv_backend="c10d",
                 max_restarts=0,
             )
-            elastic_launch(launch_config, runner_wrapper)(cfg)
+            elastic_launch(launch_config, Submitit())(cfg)
         else:
             logging.info("Running in local mode without elastic launch")
             distutils.setup_env_local()
-            runner_wrapper(cfg)
+            Submitit()(cfg)
diff --git a/src/fairchem/core/components/runner.py b/src/fairchem/core/components/runner.py
index 6f5bdfefb8..fb20246813 100644
--- a/src/fairchem/core/components/runner.py
+++ b/src/fairchem/core/components/runner.py
@@ -25,7 +25,7 @@ def job_config(self, cfg: DictConfig):
         self._job_config = cfg
 
     @abstractmethod
-    def run(self) -> Any:
+    def run(self, **kwargs) -> Any:
         raise NotImplementedError
 
     @abstractmethod

From 980ef9b3886183644c8919c112b48f2892fe29c1 Mon Sep 17 00:00:00 2001
From: lbluque <lbluque@meta.com>
Date: Wed, 12 Feb 2025 07:57:43 +1300
Subject: [PATCH 2/2] fix f string

---
 src/fairchem/core/_cli_hydra.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/fairchem/core/_cli_hydra.py b/src/fairchem/core/_cli_hydra.py
index cdf36836a6..5ccdbd5176 100644
--- a/src/fairchem/core/_cli_hydra.py
+++ b/src/fairchem/core/_cli_hydra.py
@@ -268,7 +268,7 @@ def main(
                         num_jobs=scheduler_cfg.num_jobs,
                     )
                     jobs.append(job)
-            logging.info(f"Submitted {len(jobs)} jobs: {jobs[0].job_id.split("_")[0]}")
+            logging.info(f"Submitted {len(jobs)} jobs: {jobs[0].job_id.split('_')[0]}")
     else:
         from torch.distributed.launcher.api import LaunchConfig, elastic_launch