added wandb.run_id to allow resuming without wandb log; updated log m… (#841)

Co-authored-by: Simon Alibert <75076266+aliberts@users.noreply.github.com>

lerobot/common/utils/wandb_utils.py

@@ -69,7 +69,13 @@ class WandBLogger:
         os.environ["WANDB_SILENT"] = "True"
         import wandb
 
-        wandb_run_id = get_wandb_run_id_from_filesystem(self.log_dir) if cfg.resume else None
+        wandb_run_id = (
+            cfg.wandb.run_id
+            if cfg.wandb.run_id
+            else get_wandb_run_id_from_filesystem(self.log_dir)
+            if cfg.resume
+            else None
+        )
         wandb.init(
             id=wandb_run_id,
             project=self.cfg.project,

lerobot/configs/default.py

@@ -46,6 +46,7 @@ class WandBConfig:
     project: str = "lerobot"
     entity: str | None = None
     notes: str | None = None
+    run_id: str | None = None
 
 
 @dataclass

lerobot/configs/train.py

@@ -79,7 +79,9 @@ class TrainPipelineConfig(HubMixin):
             # The entire train config is already loaded, we just need to get the checkpoint dir
             config_path = parser.parse_arg("config_path")
             if not config_path:
-                raise ValueError("A config_path is expected when resuming a run.")
+                raise ValueError(
+                    f"A config_path is expected when resuming a run. Please specify path to {TRAIN_CONFIG_NAME}"
+                )
             if not Path(config_path).resolve().exists():
                 raise NotADirectoryError(
                     f"{config_path=} is expected to be a local path. "