configuration_energy_multitask.py

from __future__ import annotations

"""Custom config for the Energy Intelligence multitask model (NER + classification)."""

from transformers import DistilBertConfig


class EnergyMultitaskConfig(DistilBertConfig):
    """Configuration for a DistilBERT model with a shared encoder and two heads:

    - **NER head**  : token-level classification (BIO entity tags)
    - **CLS head**  : sequence-level multi-label classification (topic labels)
    """

    model_type = "energy_multitask"

    def __init__(
        self,
        # --- NER head ---
        ner_num_labels: int = 19,
        ner_id2label: dict | None = None,
        ner_label2id: dict | None = None,
        # --- Classification head ---
        cls_num_labels: int = 10,
        cls_id2label: dict | None = None,
        cls_label2id: dict | None = None,
        seq_classif_dropout: float = 0.2,
        **kwargs,
    ) -> None:
        # Silence deprecation: transformers 5+ uses return_dict not use_return_dict
        kwargs.pop("use_return_dict", None)
        super().__init__(**kwargs)
        self.ner_num_labels = ner_num_labels
        self.ner_id2label = ner_id2label or {}
        self.ner_label2id = ner_label2id or {}
        self.cls_num_labels = cls_num_labels
        self.cls_id2label = cls_id2label or {}
        self.cls_label2id = cls_label2id or {}
        self.seq_classif_dropout = seq_classif_dropout