modeling_listconranker.py

# Copyright 2024 Bytedance Ltd. and/or its affiliates
#
# Permission is hereby granted, free of charge, to any person obtaining a copy of this software
# and associated documentation files (the "Software"), to deal in the Software without
# restriction, including without limitation the rights to use, copy, modify, merge, publish,
# distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the
# Software is furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all copies or
# substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
# OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
# ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
# OTHER DEALINGS IN THE SOFTWARE.
from __future__ import annotations
import torch
from torch import nn
from torch.nn import functional as F
from transformers import (
    PreTrainedModel,
    BertModel,
    AutoTokenizer,
)
import os
from transformers.modeling_outputs import SequenceClassifierOutput
from typing import Union, List, Optional
from collections import defaultdict
import numpy as np
import math
from huggingface_hub import hf_hub_download
from .configuration_listconranker import ListConRankerConfig


class QueryEmbedding(nn.Module):
    def __init__(self, config) -> None:
        super().__init__()
        self.query_embedding = nn.Embedding(2, config.list_con_hidden_size)
        self.layerNorm = nn.LayerNorm(config.list_con_hidden_size)

    def forward(self, x, tags):
        query_embeddings = self.query_embedding(tags)
        x += query_embeddings
        x = self.layerNorm(x)
        return x


class ListTransformer(nn.Module):
    def __init__(self, num_layer, config) -> None:
        super().__init__()
        self.config = config
        self.list_transformer_layer = nn.TransformerEncoderLayer(
            config.list_con_hidden_size,
            self.config.num_attention_heads,
            batch_first=True,
            activation=F.gelu,
            norm_first=False,
        )
        self.list_transformer = nn.TransformerEncoder(
            self.list_transformer_layer, num_layer
        )
        self.relu = nn.ReLU()
        self.query_embedding = QueryEmbedding(config)

        self.linear_score3 = nn.Linear(
            config.list_con_hidden_size * 2, config.list_con_hidden_size
        )
        self.linear_score2 = nn.Linear(
            config.list_con_hidden_size * 2, config.list_con_hidden_size
        )
        self.linear_score1 = nn.Linear(config.list_con_hidden_size * 2, 1)

    def forward(
        self, pair_features: torch.Tensor, pair_nums: List[int]
    ) -> torch.Tensor:
        batch_pair_features = pair_features.split(pair_nums)

        pair_feature_query_passage_concat_list = []
        for i in range(len(batch_pair_features)):
            pair_feature_query = (
                batch_pair_features[i][0].unsqueeze(0).repeat(pair_nums[i] - 1, 1)
            )
            pair_feature_passage = batch_pair_features[i][1:]
            pair_feature_query_passage_concat_list.append(
                torch.cat([pair_feature_query, pair_feature_passage], dim=1)
            )
        pair_feature_query_passage_concat = torch.cat(
            pair_feature_query_passage_concat_list, dim=0
        )

        batch_pair_features = nn.utils.rnn.pad_sequence(
            batch_pair_features, batch_first=True
        )

        query_embedding_tags = torch.zeros(
            batch_pair_features.size(0),
            batch_pair_features.size(1),
            dtype=torch.long,
            device=self.device,
        )
        query_embedding_tags[:, 0] = 1
        batch_pair_features = self.query_embedding(
            batch_pair_features, query_embedding_tags
        )

        mask = self.generate_attention_mask(pair_nums)
        query_mask = self.generate_attention_mask_custom(pair_nums)
        pair_list_features = self.list_transformer(
            batch_pair_features, src_key_padding_mask=mask, mask=query_mask
        )

        output_pair_list_features = []
        output_query_list_features = []
        pair_features_after_transformer_list = []
        for idx, pair_num in enumerate(pair_nums):
            output_pair_list_features.append(pair_list_features[idx, 1:pair_num, :])
            output_query_list_features.append(pair_list_features[idx, 0, :])
            pair_features_after_transformer_list.append(
                pair_list_features[idx, :pair_num, :]
            )

        pair_features_after_transformer_cat_query_list = []
        for idx, pair_num in enumerate(pair_nums):
            query_ft = (
                output_query_list_features[idx].unsqueeze(0).repeat(pair_num - 1, 1)
            )
            pair_features_after_transformer_cat_query = torch.cat(
                [query_ft, output_pair_list_features[idx]], dim=1
            )
            pair_features_after_transformer_cat_query_list.append(
                pair_features_after_transformer_cat_query
            )
        pair_features_after_transformer_cat_query = torch.cat(
            pair_features_after_transformer_cat_query_list, dim=0
        )

        pair_feature_query_passage_concat = self.relu(
            self.linear_score2(pair_feature_query_passage_concat)
        )
        pair_features_after_transformer_cat_query = self.relu(
            self.linear_score3(pair_features_after_transformer_cat_query)
        )
        final_ft = torch.cat(
            [
                pair_feature_query_passage_concat,
                pair_features_after_transformer_cat_query,
            ],
            dim=1,
        )
        logits = self.linear_score1(final_ft).squeeze()
        return logits, torch.cat(pair_features_after_transformer_list, dim=0)

    def generate_attention_mask(self, pair_num):
        max_len = max(pair_num)
        batch_size = len(pair_num)
        mask = torch.zeros(batch_size, max_len, dtype=torch.bool, device=self.device)
        for i, length in enumerate(pair_num):
            mask[i, length:] = True
        return mask

    def generate_attention_mask_custom(self, pair_num):
        max_len = max(pair_num)
        mask = torch.zeros(max_len, max_len, dtype=torch.bool, device=self.device)
        mask[0, 1:] = True
        return mask


class ListConRankerModel(PreTrainedModel):
    """
    ListConRanker model for sequence classification that's compatible with AutoModelForSequenceClassification.
    """

    config_class = ListConRankerConfig
    base_model_prefix = "listconranker"

    def __init__(self, config: ListConRankerConfig):
        super().__init__(config)
        self.config = config
        self.num_labels = config.num_labels
        self.hf_model = BertModel(config.bert_config)

        self.sigmoid = nn.Sigmoid()

        self.linear_in_embedding = nn.Linear(
            config.hidden_size, config.list_con_hidden_size
        )
        self.list_transformer = ListTransformer(
            config.list_transformer_layers,
            config,
        )

    def forward(
        self,
        input_ids: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        **kwargs,
    ) -> Union[tuple[torch.Tensor], SequenceClassifierOutput]:
        if self.training:
            raise NotImplementedError("Training not supported; use eval mode.")
        device = input_ids.device
        self.list_transformer.device = device
        # Reorganize by unique queries and their passages
        (
            reorganized_input_ids,
            reorganized_attention_mask,
            reorganized_token_type_ids,
            pair_nums,
            group_indices,
        ) = self._reorganize_inputs(input_ids, attention_mask, token_type_ids)

        out = self.hf_model(
            input_ids=reorganized_input_ids,
            attention_mask=reorganized_attention_mask,
            token_type_ids=reorganized_token_type_ids,
            return_dict=True,
        )
        feats = out.last_hidden_state
        pooled = self.average_pooling(feats, reorganized_attention_mask)
        embedded = self.linear_in_embedding(pooled)
        logits, _ = self.list_transformer(embedded, pair_nums)
        probs = self.sigmoid(logits)

        # Restore original order
        sorted_probs = self._restore_original_order(probs, group_indices)
        sorted_logits = self._restore_original_order(logits, group_indices)
        if not return_dict:
            return (sorted_probs, sorted_logits)

        return SequenceClassifierOutput(
            loss=None,
            logits=sorted_logits,
            hidden_states=out.hidden_states,
            attentions=out.attentions,
        )

    def _reorganize_inputs(
        self,
        input_ids: torch.Tensor,
        attention_mask: torch.Tensor,
        token_type_ids: Optional[torch.Tensor],
    ) -> tuple[
        torch.Tensor, torch.Tensor, Optional[torch.Tensor], List[int], List[List[int]]
    ]:
        """
        Group inputs by unique queries: for each query, produce [query] + its passages,
        then flatten, pad, and return pair sizes and original indices mapping.
        """
        batch_size = input_ids.size(0)
        # Structure: query_key -> {
        #   'query': (seq, mask, tt),
        #   'passages': [(seq, mask, tt), ...],
        #   'indices': [original_index, ...]
        # }
        grouped = {}

        for idx in range(batch_size):
            seq = input_ids[idx]
            mask = attention_mask[idx]
            token_type_ids[idx] if token_type_ids is not None else torch.zeros_like(seq)

            sep_idxs = (seq == self.config.sep_token_id).nonzero(as_tuple=True)[0]
            if sep_idxs.numel() == 0:
                raise ValueError(f"No SEP in sequence {idx}")
            first_sep = sep_idxs[0].item()
            second_sep = sep_idxs[1].item()

            # Extract query and passage
            q_seq = seq[: first_sep + 1]
            q_mask = mask[: first_sep + 1]
            q_tt = torch.zeros_like(q_seq)

            p_seq = seq[first_sep : second_sep + 1]
            p_mask = mask[first_sep : second_sep + 1]
            p_seq = p_seq.clone()
            p_seq[0] = self.config.cls_token_id
            p_tt = torch.zeros_like(p_seq)

            # Build key excluding CLS/SEP
            key = tuple(
                q_seq[
                    (q_seq != self.config.cls_token_id)
                    & (q_seq != self.config.sep_token_id)
                ].tolist()
            )

            # truncation
            q_seq = q_seq[: self.config.max_position_embeddings]
            q_seq[-1] = self.config.sep_token_id
            p_seq = p_seq[: self.config.max_position_embeddings]
            p_seq[-1] = self.config.sep_token_id
            q_mask = q_mask[: self.config.max_position_embeddings]
            p_mask = p_mask[: self.config.max_position_embeddings]
            q_tt = q_tt[: self.config.max_position_embeddings]
            p_tt = p_tt[: self.config.max_position_embeddings]

            if key not in grouped:
                grouped[key] = {
                    "query": (q_seq, q_mask, q_tt),
                    "passages": [],
                    "indices": [],
                }
            grouped[key]["passages"].append((p_seq, p_mask, p_tt))
            grouped[key]["indices"].append(idx)

        # Flatten according to group insertion order
        seqs, masks, tts, pair_nums, group_indices = [], [], [], [], []
        for key, data in grouped.items():
            q_seq, q_mask, q_tt = data["query"]
            passages = data["passages"]
            indices = data["indices"]
            # record sizes and original positions
            pair_nums.append(len(passages) + 1)  # +1 for the query
            group_indices.append(indices)

            # append query then its passages
            seqs.append(q_seq)
            masks.append(q_mask)
            tts.append(q_tt)
            for p_seq, p_mask, p_tt in passages:
                seqs.append(p_seq)
                masks.append(p_mask)
                tts.append(p_tt)

        # Pad to uniform length
        max_len = max(s.size(0) for s in seqs)
        padded_seqs, padded_masks, padded_tts = [], [], []
        for s, m, t in zip(seqs, masks, tts):
            ps = torch.zeros(max_len, dtype=s.dtype, device=s.device)
            pm = torch.zeros(max_len, dtype=m.dtype, device=m.device)
            pt = torch.zeros(max_len, dtype=t.dtype, device=t.device)
            ps[: s.size(0)] = s
            pm[: m.size(0)] = m
            pt[: t.size(0)] = t
            padded_seqs.append(ps)
            padded_masks.append(pm)
            padded_tts.append(pt)

        rid = torch.stack(padded_seqs)
        ram = torch.stack(padded_masks)
        rtt = torch.stack(padded_tts) if token_type_ids is not None else None

        return rid, ram, rtt, pair_nums, group_indices

    def _restore_original_order(
        self,
        logits: torch.Tensor,
        group_indices: List[List[int]],
    ) -> torch.Tensor:
        """
        Map flattened logits back so each original index gets its passage score.
        """
        out = torch.zeros(logits.size(0), dtype=logits.dtype, device=logits.device)
        i = 0
        for indices in group_indices:
            for idx in indices:
                out[idx] = logits[i]
                i += 1
        return out.reshape(-1, 1)

    def average_pooling(self, hidden_state, attention_mask):
        extended_attention_mask = (
            attention_mask.unsqueeze(-1)
            .expand(hidden_state.size())
            .to(dtype=hidden_state.dtype)
        )
        masked_hidden_state = hidden_state * extended_attention_mask
        sum_embeddings = torch.sum(masked_hidden_state, dim=1)
        sum_mask = extended_attention_mask.sum(dim=1)
        return sum_embeddings / sum_mask

    @classmethod
    def from_pretrained(
        cls, model_name_or_path, config: Optional[ListConRankerConfig] = None, **kwargs
    ):
        model = super().from_pretrained(model_name_or_path, config=config, **kwargs)
        model.hf_model = BertModel.from_pretrained(
            model_name_or_path, config=model.config.bert_config, **kwargs
        )
        linear_path = hf_hub_download(
            repo_id = model_name_or_path,
            filename = "linear_in_embedding.pt",
            revision = "main",
            cache_dir = kwargs['cache_dir'] if 'cache_dir' in kwargs else None
        )
        list_transformer_path = hf_hub_download(
            repo_id = "ByteDance/ListConRanker",
            filename = "list_transformer.pt",
            revision = "main",
            cache_dir = kwargs['cache_dir'] if 'cache_dir' in kwargs else None
        )

        try:
            model.linear_in_embedding.load_state_dict(torch.load(linear_path))
            model.list_transformer.load_state_dict(torch.load(list_transformer_path))
        except FileNotFoundError as e:
            raise e

        return model

    def multi_passage(
        self,
        sentences: List[List[str]],
        batch_size: int = 32,
        tokenizer: AutoTokenizer = AutoTokenizer.from_pretrained(
            "ByteDance/ListConRanker"
        ),
    ):
        """
        Process multiple passages for each query.
        :param sentences: List of lists, where each inner list contains sentences for a query.
        :return: Tensor of logits for each passage.
        """
        pairs = []
        for batch in sentences:
            if len(batch) < 2:
                raise ValueError("Each query must have at least one passage.")
            query = batch[0]
            passages = batch[1:]
            for passage in passages:
                pairs.append((query, passage))

        total_batches = (len(pairs) + batch_size - 1) // batch_size
        total_logits = torch.zeros(len(pairs), dtype=torch.float, device=self.device)
        for batch in range(total_batches):
            batch_pairs = pairs[batch * batch_size : (batch + 1) * batch_size]
            inputs = tokenizer(
                batch_pairs,
                padding=True,
                truncation=False,
                return_tensors="pt",
            )

            for k, v in inputs.items():
                inputs[k] = v.to(self.device)

            logits = self(**inputs)[0]
            total_logits[batch * batch_size : (batch + 1) * batch_size] = (
                logits.squeeze(1)
            )
        return total_logits.tolist()

    def multi_passage_in_iterative_inference(
        self,
        sentences: List[str],
        stop_num: int = 20,
        decrement_rate: float = 0.2,
        min_filter_num: int = 10,
        tokenizer: AutoTokenizer = AutoTokenizer.from_pretrained(
            "ByteDance/ListConRanker"
        ),
    ):
        """
        Process multiple passages for one query in iterative inference.
        :param sentences: List contains sentences for a query.
        :return: Tensor of logits for each passage.
        """
        if stop_num < 1:
            raise ValueError("stop_num must be greater than 0")
        if decrement_rate <= 0 or decrement_rate >= 1:
            raise ValueError("decrement_rate must be in (0, 1)")
        if min_filter_num < 1:
            raise ValueError("min_filter_num must be greater than 0")

        query = sentences[0]
        passage = sentences[1:]

        filter_times = 0
        passage2score = defaultdict(list)
        while len(passage) > stop_num:
            batch = [[query] + passage]
            pred_scores = self.multi_passage(
                batch, batch_size=len(batch[0]) - 1, tokenizer=tokenizer
            )
            pred_scores_argsort = np.argsort(
                pred_scores
            ).tolist()  # Sort in increasing order

            passage_len = len(passage)
            to_filter_num = math.ceil(passage_len * decrement_rate)
            if to_filter_num < min_filter_num:
                to_filter_num = min_filter_num

            have_filter_num = 0
            while have_filter_num < to_filter_num:
                idx = pred_scores_argsort[have_filter_num]
                passage2score[passage[idx]].append(pred_scores[idx] + filter_times)
                have_filter_num += 1
            while (
                pred_scores[pred_scores_argsort[have_filter_num - 1]]
                == pred_scores[pred_scores_argsort[have_filter_num]]
            ):
                idx = pred_scores_argsort[have_filter_num]
                passage2score[passage[idx]].append(pred_scores[idx] + filter_times)
                have_filter_num += 1
            next_passage = []
            next_passage_idx = have_filter_num
            while next_passage_idx < len(passage):
                idx = pred_scores_argsort[next_passage_idx]
                next_passage.append(passage[idx])
                next_passage_idx += 1
            passage = next_passage
            filter_times += 1

        batch = [[query] + passage]
        pred_scores = self.multi_passage(
            batch, batch_size=len(batch[0]) - 1, tokenizer=tokenizer
        )

        cnt = 0
        while cnt < len(passage):
            passage2score[passage[cnt]].append(pred_scores[cnt] + filter_times)
            cnt += 1

        passage = sentences[1:]
        final_score = []
        for i in range(len(passage)):
            p = passage[i]
            final_score.append(passage2score[p][0])
        return final_score