Old code clean-up

PAD_ID = 0

class ModelOutput(NamedTuple):

    pred_binary: torch.Tensor

    pred_binary_final: Optional[torch.Tensor] = None

    pred_regression: Optional[torch.Tensor] = None

    pred_multilabel: Optional[torch.Tensor] = None

    attn_binary: Optional[torch.Tensor] = None

    attn_regression: Optional[torch.Tensor] = None

    word_attns: Optional[torch.Tensor] = None

    word_conicity: Optional[torch.Tensor] = None

    sent_conicity: Optional[torch.Tensor] = None

    chunk_hidden_states: Optional[torch.Tensor] = None

class PHQTotalMulticlassAttentionModelBERT(nn.Module):

    """Hierarchical Attention Classification model with BERT in the word level.

        self,

        inputs: Dict[str, torch.Tensor],

        return_attn: bool = False,

    ) -> ModelOutput:

    ) -> torch.Tensor:

        model_output = self.encoder(inputs["input_ids"], inputs["attention_mask"])

        sentence_embeddings = self.mean_pooling(model_output, inputs["attention_mask"])

        word_outputs = torch.split(sentence_embeddings, inputs["text_lens"].tolist())

import json

import itertools

import json

from pathlib import Path

from typing import Dict, List, Optional