a few more clean ups

Author: allenwang28

apps/grpo/main.py

@@ -26,20 +26,13 @@
 from forge.actors.trainer import RLTrainer
 from forge.cli.config import parse
 from forge.controller.actor import ForgeActor
-from forge.controller.launcher import JOB_NAME_KEY, LAUNCHER_KEY
 from forge.controller.provisioner import init_provisioner, shutdown
 from forge.data.rewards import MathReward, ThinkingReward
 from forge.observability.metric_actors import get_or_create_metric_logger
 from forge.observability.metrics import record_metric, Reduce
 from forge.observability.perf_tracker import Tracer
 
-from forge.types import (
-    Launcher,
-    LauncherConfig,
-    ProcessConfig,
-    ProvisionerConfig,
-    ServiceConfig,
-)
+from forge.types import LauncherConfig, ProvisionerConfig
 from forge.util.ops import compute_logprobs
 from monarch.actor import endpoint
 from omegaconf import DictConfig
@@ -320,25 +313,21 @@ async def main(cfg: DictConfig):
     max_req_tokens = cfg.max_req_tokens
     max_res_tokens = cfg.max_res_tokens
 
-    # init provisioner
-    await init_provisioner(
-        ProvisionerConfig(
-            launcher_config=LauncherConfig(
-                launcher=Launcher(cfg.get(LAUNCHER_KEY, Launcher.SLURM.value)),
-                job_name=cfg.get(JOB_NAME_KEY, None),
-                services={k: ServiceConfig(**v) for k, v in cfg.services.items()},
-                actors={k: ProcessConfig(**v) for k, v in cfg.actors.items()},
+    # ---- Global setups ---- #
+    if cfg.get("provisioner", None) is not None:
+        await init_provisioner(
+            ProvisionerConfig(
+                launcher_config=LauncherConfig(**cfg.provisioner.launcher)
             )
         )
-    )
-
     # initialize before spawning services
     metric_logging_cfg = cfg.get("metric_logging", {"console": {"log_per_rank": False}})
     mlogger = await get_or_create_metric_logger()
     await mlogger.init_backends.call_one(metric_logging_cfg)
+    await ts.initialize(strategy=ts.ControllerStorageVolumes())
 
     # ---- Setup services ---- #
-    await ts.initialize(strategy=ts.ControllerStorageVolumes())
+
     (
         dataloader,
         policy,

apps/grpo/qwen3_32b.yaml

@@ -10,6 +10,9 @@ max_res_tokens: 512
 model: "Qwen/Qwen3-32B"
 off_by_n: 1 # Off by one by default
 
+provisioner:
+  launcher: slurm
+
 # Main loop configuration
 rollout_threads: 1   # Recommended to set equal to policy.num_replicas
 

src/forge/controller/provisioner.py

@@ -36,20 +36,20 @@ def _get_port() -> str:
         return str(port)
 
 
-class _SetupActor(Actor):
+class _RemoteInfoFetcher(Actor):
     """An actor responsible for getting remote host information."""
 
     @endpoint
-    def get_info(self) -> [str, str]:
+    def get_info(self) -> tuple[str, str]:
         return socket.gethostname(), _get_port()
 
 
 async def get_remote_info(host_mesh: HostMesh) -> tuple[str, str]:
     """Returns the host name and port of the host mesh."""
     throwaway_procs = host_mesh.spawn_procs(per_host={"procs": 1})
-    setup_actor = throwaway_procs.spawn("_setup_actor", _SetupActor)
-    setup_actor = setup_actor.slice(procs=0)
-    host, port = await setup_actor.get_info.call_one()
+    fetcher = throwaway_procs.spawn("_fetcher", _RemoteInfoFetcher)
+    fetcher = fetcher.slice(procs=0)
+    host, port = await fetcher.get_info.call_one()
     await throwaway_procs.stop()
     return host, port