Minor fix (suryadheeshjith#5)

Fixes to README and logging in train during checkpoint saving.

Author: suryadheeshjith

README.md

@@ -27,6 +27,7 @@ cd Samudra
 Using conda:
 ```bash
 conda env create -f environment.yml
+conda activate samudra
 ```
 
 Using [`uv`](https://docs.astral.sh/uv/):
@@ -47,7 +48,7 @@ A default training configuration is provided in the file configs/train_samudra_o
 > Note: Ensure your environment is activated before training.
 ```bash
 # Train a new model
-python src/train.py --config path/to/train_config.yaml
+torchrun src/train.py --config path/to/train_config.yaml
 ```
 
 ### Rollout

src/train.py

@@ -533,8 +533,8 @@ def save_all_checkpoints(self, epoch, v_loss, inf_loss):
         save_best_checkpoint = False
         if v_loss <= self.best_val_loss:
             logging.info(
-                f"Epoch validation loss ({v_loss}) is lower than "
-                f"previous best validation loss ({self.best_val_loss})."
+                f"Epoch validation loss ({v_loss:.3f}) is lower than "
+                f"previous best validation loss ({self.best_val_loss:.3f})."
             )
             logging.info(
                 "Saving lowest validation loss checkpoint to "
@@ -544,8 +544,8 @@ def save_all_checkpoints(self, epoch, v_loss, inf_loss):
             save_best_checkpoint = True  # wait until inference error is updated
         if inf_loss is not None and (inf_loss <= self.best_inf_loss):
             logging.info(
-                f"Epoch inference error ({inf_loss}) is lower than "
-                f"previous best inference error ({self.best_inf_loss})."
+                f"Epoch inference error ({inf_loss:.3f}) is lower than "
+                f"previous best inference error ({self.best_inf_loss:.3f})."
             )
             logging.info(
                 "Saving lowest inference error checkpoint to "