use tuples for weighting loss functions to support yaml

lightning_ir/base/module.py

@@ -2,7 +2,7 @@
 
 from collections import defaultdict
 from pathlib import Path
-from typing import TYPE_CHECKING, Any, Dict, Mapping, Sequence
+from typing import TYPE_CHECKING, Any, Dict, Sequence, Tuple
 
 import torch
 from lightning import LightningModule
@@ -22,7 +22,7 @@ def __init__(
         model_name_or_path: str | None = None,
         config: LightningIRConfig | None = None,
         model: LightningIRModel | None = None,
-        loss_functions: Sequence[LossFunction] | Mapping[LossFunction, float] | None = None,
+        loss_functions: Sequence[LossFunction] | Sequence[Tuple[LossFunction, float]] | None = None,
         evaluation_metrics: Sequence[str] | None = None,
     ):
         super().__init__()
@@ -43,9 +43,14 @@ def __init__(
 
         self.model: LightningIRModel = model
         self.config = self.model.config
-        if loss_functions is not None and not isinstance(loss_functions, dict):
-            loss_functions = {loss_function: 1.0 for loss_function in loss_functions}
-        self.loss_functions = loss_functions
+        self.loss_functions: Dict[LossFunction, float] | None = None
+        if loss_functions is not None:
+            self.loss_functions = {}
+            for loss_function in loss_functions:
+                if isinstance(loss_function, LossFunction):
+                    self.loss_functions[loss_function] = 1.0
+                else:
+                    self.loss_functions[loss_function[0]] = loss_function[1]
         self.evaluation_metrics = evaluation_metrics
         self.tokenizer = self.config.__class__.tokenizer_class.from_pretrained(
             self.config.name_or_path, **self.config.to_tokenizer_dict()