File size: 9,293 Bytes
714cf46
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
import itertools
import logging
from collections import defaultdict
from collections.abc import Sequence
from typing import Iterator, TypedDict
import torch
from tqdm import tqdm
from .FastPLMs.e1.modeling_e1 import E1ForMaskedLM, E1MaskedLMOutputWithPast, E1BatchPreparer, get_context, DataPrepConfig, KVCache

IndexedSequence = tuple[int, str]
logger = logging.getLogger(__name__)


class E1Prediction(TypedDict, total=False):
    id: str | int
    context_id: str | int | None
    logits: torch.Tensor
    token_embeddings: torch.Tensor
    mean_token_embeddings: torch.Tensor


class E1Predictor:
    def __init__(
        self,
        model: E1ForMaskedLM,
        data_prep_config: DataPrepConfig | None = None,
        max_batch_tokens: int = 65536,
        use_cache: bool = True,
        cache_size: int = 4,
        save_masked_positions_only: bool = False,
        fields_to_save: list[str] = ["logits", "token_embeddings", "mean_token_embeddings"],
        keep_predictions_in_gpu: bool = False,
    ):
        self.model = model
        self.max_batch_tokens = max_batch_tokens
        self.batch_preparer = E1BatchPreparer(data_prep_config=data_prep_config)
        self.model.eval()
        self.kv_cache = KVCache(cache_size=cache_size) if use_cache else None

        self.fields_to_save = fields_to_save
        self.save_masked_positions_only = save_masked_positions_only
        self.keep_predictions_in_gpu = keep_predictions_in_gpu

    def group_by_length(self, indexed_sequences: list[IndexedSequence]) -> list[list[IndexedSequence]]:
        batches: list[list[IndexedSequence]] = [[]]
        for idx, seq in sorted(indexed_sequences, key=lambda idx_seq: (len(idx_seq[1]), idx_seq[0])):
            if len(batches[-1]) > 0 and len(seq) * (len(batches[-1]) + 1) > self.max_batch_tokens:
                batches.append([])
            batches[-1].append((idx, seq))

        return batches

    def group_by_context(self, indexed_sequences: list[IndexedSequence]) -> list[list[IndexedSequence]]:
        batches: dict[str | None, list[IndexedSequence]] = defaultdict(list)
        for idx, seq in indexed_sequences:
            batches[get_context(seq)].append((idx, seq))
        return list(batches.values())

    def batch_sequences(self, sequences: list[str]) -> list[tuple[list[int], bool]]:  # type: ignore[override]
        """
        Batches the sequences and returns indices for the current rank
        We want to keep sequences of similar length together.
        Ensures that no batch exceeds max_batch_tokens
        [For E1, also ensures if context is present, preserve locality of context]
        """
        indexed_sequences: list[IndexedSequence] = list(enumerate(sequences))
        indexed_batches = self.group_by_context(indexed_sequences)
        # Preserve context ordering
        indexed_batches = list(
            itertools.chain.from_iterable([self.group_by_length(batch) for batch in indexed_batches])
        )
        batches = [[item[0] for item in batch] for batch in indexed_batches]  # type: ignore[no-redef,misc]

        assert sorted(sum(batches, [])) == list(range(len(sequences))), (
            "Batches must contain all indices with no repetition"
        )

        batches_with_validity = [(b, True) for b in batches]

        return batches_with_validity

    @torch.no_grad()
    def predict_batch(self, sequences: list[str], sequence_metadata: list[dict[str, str | int]]) -> list[E1Prediction]:
        """
        Returns the logits/embeddings for the last sequence for multi-sequence inputs.
        """
        outputs = self.predict_batch_padded(sequences)
        outputs["logits"] = outputs["logits"].float()
        outputs["embeddings"] = outputs["embeddings"].float()

        token_mask = outputs["non_boundary_token_mask"] & outputs["last_sequence_mask"]
        if self.save_masked_positions_only:
            token_mask = token_mask & outputs["mask_positions_mask"]
        predictions = []
        for i in range(len(sequences)):
            pred: E1Prediction = {
                "id": sequence_metadata[i]["id"],
                "context_id": sequence_metadata[i].get("context_id", None),
            }
            if "logits" in self.fields_to_save:
                pred["logits"] = outputs["logits"][i, token_mask[i]]
                if not self.keep_predictions_in_gpu:
                    pred["logits"] = pred["logits"].to("cpu")  # type: ignore[union-attr]
            if "token_embeddings" in self.fields_to_save:
                pred["token_embeddings"] = outputs["embeddings"][i, token_mask[i]]
                if not self.keep_predictions_in_gpu:
                    pred["token_embeddings"] = pred["token_embeddings"].to("cpu")  # type: ignore[union-attr]
            if "mean_token_embeddings" in self.fields_to_save:
                pred["mean_token_embeddings"] = outputs["embeddings"][i, token_mask[i]].mean(dim=0)
                if not self.keep_predictions_in_gpu:
                    pred["mean_token_embeddings"] = pred["mean_token_embeddings"].to("cpu")  # type: ignore[union-attr]
            predictions.append(pred)
        return predictions

    @torch.no_grad()
    def predict_batch_padded(self, sequences: list[str]) -> dict[str, torch.Tensor]:
        """
        If use_cache is True, this function will return the logits/embeddings for the last sequence for multi-sequence inputs.
        If use_cache is False, this function will return the logits/embeddings for every sequence for multi-sequence inputs.

        Returns three additional masks:
        - non_boundary_token_mask: True for tokens that are part of the input sequence i.e not boundary tokens like 1, 2, <bos>, <eos>, <pad>, etc.
        - last_sequence_mask: True for tokens that are part of the last sequence (including boundary tokens) in case of multi-sequence input.
        - mask_positions_mask: True for masked positions.
        - valid_token_mask: True for valid tokens.
        """
        device_type = "cuda" if torch.cuda.is_available() else "cpu"
        with torch.autocast(device_type, torch.bfloat16):
            batch = self.batch_preparer.get_batch_kwargs(sequences, device=torch.device(device_type))

            if self.kv_cache is not None:
                self.kv_cache.before_forward(batch)

            output: E1MaskedLMOutputWithPast = self.model(
                input_ids=batch["input_ids"],
                within_seq_position_ids=batch["within_seq_position_ids"],
                global_position_ids=batch["global_position_ids"],
                sequence_ids=batch["sequence_ids"],
                past_key_values=batch.get("past_key_values", None),
                use_cache=batch.get("use_cache", False),
                output_attentions=False,
                output_hidden_states=False,
            )
            if self.kv_cache is not None:
                self.kv_cache.after_forward(batch, output)

            logits = output.logits
            embeddings = output.last_hidden_state

            padding_mask = batch["input_ids"] == self.batch_preparer.pad_token_id
            last_sequence_mask = batch["sequence_ids"] == batch["sequence_ids"].max(dim=1)[0][:, None]  # type: ignore[union-attr]
            boundary_token_mask = self.batch_preparer.get_boundary_token_mask(batch["input_ids"])
            mask_positions_mask = self.batch_preparer.get_mask_positions_mask(batch["input_ids"])

            return {
                "logits": logits,
                "embeddings": embeddings,
                "last_sequence_mask": last_sequence_mask,
                "non_boundary_token_mask": ~boundary_token_mask,
                "mask_positions_mask": mask_positions_mask,
                "valid_token_mask": ~padding_mask,
            }

    @torch.no_grad()
    def predict(
        self,
        sequences: Sequence[str],
        sequence_ids: Sequence[int | str] | None = None,
        context_seqs: dict[str, str] | None = None,
    ) -> Iterator[E1Prediction]:
        if sequence_ids is None:
            sequence_ids = list(range(len(sequences)))
        if context_seqs:
            sequences_with_context = [
                (ctx + "," + seq, {"context_id": ctx_id, "id": sequence_id})
                for ctx_id, ctx in context_seqs.items()
                for seq, sequence_id in zip(sequences, sequence_ids)
            ]
        else:
            sequences_with_context = [(seq, {"id": sequence_id}) for seq, sequence_id in zip(sequences, sequence_ids)]
        sequences, sequence_metadata = tuple(zip(*sequences_with_context))  # type: ignore[assignment]
        sequence_batch_indices: list[tuple[list[int], bool]] = self.batch_sequences(sequences)  # type: ignore[arg-type]
        logger.info(f"Predicting for {len(sequence_batch_indices)} batches")

        for indices, is_valid_batch in tqdm(
            sequence_batch_indices, desc="Predicting batches"
        ):
            sequence_batch = [sequences[i] for i in indices]
            sequence_batch_metadata = [sequence_metadata[i] for i in indices]
            batch_predictions = self.predict_batch(sequence_batch, sequence_batch_metadata)

            if not is_valid_batch:
                continue

            for prediction in batch_predictions:
                yield prediction