Adding support of domains to config

Author: paultltc

src/nanotron/config/config.py

@@ -111,9 +111,12 @@ def __post_init__(self):
 class MultilingualNanosetDatasetsArgs:
     training_folder: Union[str, dict, List[str]]
     validation_folder: Union[str, List[str]]
-    languages: List[str]  # NOTE(tj.solergibert) Required for 1. Aggregating the result 2. Reporting to WANDB
+    domains: Optional[List[str]] = None  # NOTE(tj.solergibert) Required for 1. Aggregating the result 2. Reporting to WANDB
+    languages: Optional[List[str]] = None   # NOTE(@paultltc): For back-compatibility
 
     def __post_init__(self):
+        if self.languages is not None and self.domains is None:
+            self.domains = self.languages
         if isinstance(self.training_folder, str):  # Case 1: 1 Dataset folder
             self.training_folder = [self.training_folder]
             self.validation_folder = [self.validation_folder]
@@ -125,13 +128,13 @@ def __post_init__(self):
             self.training_folder = list(tmp_training_folder.keys())
             self.dataset_weights = list(tmp_training_folder.values())
 
-        assert len(self.training_folder) == len(
-            self.languages
-        ), f"The sizes of training_folder and languages mismatch ({len(self.training_folder)} vs {len(self.languages)})"
+        # assert len(self.training_folder) == len(
+        #     self.domains
+        # ), f"The sizes of training_folder and domains mismatch ({len(self.training_folder)} vs {len(self.domains)})"
 
-        assert len(self.training_folder) == len(
-            self.validation_folder
-        ), f"The sizes of training_folder and validation_folder mismatch ({len(self.training_folder)} vs {len(self.validation_folder)})"
+        # assert len(self.training_folder) == len(
+        #     self.validation_folder
+        # ), f"The sizes of training_folder and validation_folder mismatch ({len(self.training_folder)} vs {len(self.validation_folder)})"
 
 
 @dataclass
@@ -189,7 +192,6 @@ class GeneralArgs:
 
     Args:
         project: Name of the project (a project gather several runs in common tensorboard/hub-folders)
-        entity: Weights and bias entity name (optional)
         run: Name of the run
         step: Global step (updated when we save the checkpoint)
         consumed_train_samples: Number of samples consumed during training (should be actually just step*batch_size)

src/nanotron/trainer.py

@@ -679,7 +679,7 @@ def validation_step(self, dataloader: Iterator[Dict[str, Union[torch.Tensor, Ten
         )
 
         lang_losses = {
-            lang: [] for lang in self.config.data_stages[self.metadata.last_stage_idx].data.dataset.languages
+            lang: [] for lang in self.config.data_stages[self.metadata.last_stage_idx].data.dataset.domains
         }
         lang_losses_list = list(lang_losses.keys())