File size: 8,817 Bytes
735107c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
726b567
735107c
 
 
 
 
 
 
 
 
 
 
 
 
b05a1ce
735107c
 
 
 
 
2d000c7
 
735107c
 
 
 
 
 
 
9ea7bd1
735107c
 
 
b05a1ce
735107c
 
7cddc19
 
 
 
735107c
 
 
9ea7bd1
735107c
 
 
b05a1ce
735107c
 
 
726b567
735107c
726b567
735107c
 
 
 
34c0430
7fcb3e9
34c0430
9ea7bd1
735107c
 
 
 
 
b05a1ce
735107c
 
 
 
 
0b4bc5a
2d000c7
0b4bc5a
735107c
 
 
 
 
 
 
2d000c7
 
 
 
 
735107c
 
 
 
 
 
 
 
 
 
 
 
 
0b4bc5a
735107c
 
 
 
0b4bc5a
735107c
0b4bc5a
735107c
 
b05a1ce
735107c
 
0b4bc5a
735107c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
726b567
735107c
 
 
 
 
 
 
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
# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Perplexity Metric."""

import datasets
import numpy as np
import torch
from torch.nn import CrossEntropyLoss
from transformers import AutoModelForCausalLM, AutoTokenizer

import evaluate
from evaluate import logging


_CITATION = """\

"""

_DESCRIPTION = """
Perplexity (PPL) is one of the most common metrics for evaluating language models.
It is defined as the exponentiated average negative log-likelihood of a sequence, calculated with exponent base `e`.

For more information, see https://huggingface.co/docs/transformers/perplexity
"""

_KWARGS_DESCRIPTION = """
Args:
    model_id (str): model used for calculating Perplexity
            NOTE: Perplexity can only be calculated for causal language models.
                    This includes models such as gpt2, causal variations of bert,
                    causal versions of t5, and more (the full list can be found
                    in the AutoModelForCausalLM documentation here:
                    https://huggingface.co/docs/transformers/master/en/model_doc/auto#transformers.AutoModelForCausalLM )

    predictions (list of str): input text, each separate text snippet
        is one list entry.
    batch_size (int): the batch size to run texts through the model. Defaults to 16.
    add_start_token (bool): whether to add the start token to the texts,
        so the perplexity can include the probability of the first word. Defaults to True.
    device (str): device to run on, defaults to 'cuda' when available
    model (AutoModelForCausalLM): the model for calculating Perplexity, if provided, the model won't initialized from model_id
    tokenizer (AutoTokenizer): the tokenizer for calculating Perplexity, if provided, the tokenizer won't initialized from model_id
Returns:
    perplexity: dictionary containing the perplexity scores for the texts
        in the input list, as well as the mean perplexity. If one of the input texts is
        longer than the max input length of the model, then it is truncated to the
        max length for the perplexity computation.
Examples:
    Example 1:
        >>> perplexity = evaluate.load("perplexity", module_type="metric")
        >>> input_texts = ["lorem ipsum", "Happy Birthday!", "Bienvenue"]
        >>> results = perplexity.compute(model_id='gpt2',
        ...                              add_start_token=False,
        ...                              predictions=input_texts) # doctest:+ELLIPSIS
        >>> print(list(results.keys()))
        ['perplexities', 'mean_perplexity']
        >>> print(round(results["mean_perplexity"], 0))
        647.0
        >>> print(round(results["perplexities"][0], 0))
        32.0

    Example 2:
        >>> from datasets import load_dataset
        >>> perplexity = evaluate.load("perplexity", module_type="metric")
        >>> input_texts = load_dataset("wikitext", "wikitext-2-raw-v1", split="test")["text"][:10] # doctest: +SKIP
        >>> input_texts = [s for s in input_texts if s!='']
        >>> results = perplexity.compute(model_id='gpt2',
        ...                              predictions=input_texts)
        >>> print(list(results.keys()))
        ['perplexities', 'mean_perplexity']
        >>> print(round(results["mean_perplexity"], 2)) # doctest: +SKIP
        576.76
        >>> print(round(results["perplexities"][0], 2)) # doctest: +SKIP
        889.28
"""


@evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
class Perplexity(evaluate.Metric):
    def _info(self):
        return evaluate.MetricInfo(
            module_type="metric",
            description=_DESCRIPTION,
            citation=_CITATION,
            inputs_description=_KWARGS_DESCRIPTION,
            features=datasets.Features(
                {
                    "predictions": datasets.Value("string"),
                }
            ),
            reference_urls=["https://huggingface.co/docs/transformers/perplexity"],
        )

    def _compute(
        self, predictions, model_id, batch_size: int = 16, add_start_token: bool = True, device=None, max_length=None, model = None, tokenizer = None
    ):

        if device is not None:
            assert device in ["gpu", "cpu", "cuda"], "device should be either gpu or cpu."
            if device == "gpu":
                device = "cuda"
        else:
            device = "cuda" if torch.cuda.is_available() else "cpu"
        if model is None:
            model = AutoModelForCausalLM.from_pretrained(model_id)
            model = model.to(device)
        if tokenizer is None:
            tokenizer = AutoTokenizer.from_pretrained(model_id)

        # if batch_size > 1 (which generally leads to padding being required), and
        # if there is not an already assigned pad_token, assign an existing
        # special token to also be the padding token
        if tokenizer.pad_token is None and batch_size > 1:
            existing_special_tokens = list(tokenizer.special_tokens_map_extended.values())
            # check that the model already has at least one special token defined
            assert (
                len(existing_special_tokens) > 0
            ), "If batch_size > 1, model must have at least one special token to use for padding. Please use a different model or set batch_size=1."
            # assign one of the special tokens to also be the pad token
            tokenizer.add_special_tokens({"pad_token": existing_special_tokens[0]})

        if add_start_token and max_length:
            # leave room for <BOS> token to be added:
            assert (
                tokenizer.bos_token is not None
            ), "Input model must already have a BOS token if using add_start_token=True. Please use a different model, or set add_start_token=False"
            max_tokenized_len = max_length - 1
        else:
            max_tokenized_len = max_length

        encodings = tokenizer(
            predictions,
            add_special_tokens=False,
            padding=True,
            truncation=True if max_tokenized_len else False,
            max_length=max_tokenized_len,
            return_tensors="pt",
            return_attention_mask=True,
        ).to(device)

        encoded_texts = encodings["input_ids"]
        attn_masks = encodings["attention_mask"]

        # check that each input is long enough:
        if add_start_token:
            assert torch.all(torch.ge(attn_masks.sum(1), 1)), "Each input text must be at least one token long."
        else:
            assert torch.all(
                torch.ge(attn_masks.sum(1), 2)
            ), "When add_start_token=False, each input text must be at least two tokens long. Run with add_start_token=True if inputting strings of only one token, and remove all empty input strings."

        ppls = []
        loss_fct = CrossEntropyLoss(reduction="none")

        for start_index in logging.tqdm(range(0, len(encoded_texts), batch_size)):
            end_index = min(start_index + batch_size, len(encoded_texts))
            encoded_batch = encoded_texts[start_index:end_index]
            attn_mask = attn_masks[start_index:end_index]

            if add_start_token:
                bos_tokens_tensor = torch.tensor([[tokenizer.bos_token_id]] * encoded_batch.size(dim=0)).to(device)
                encoded_batch = torch.cat([bos_tokens_tensor, encoded_batch], dim=1)
                attn_mask = torch.cat(
                    [torch.ones(bos_tokens_tensor.size(), dtype=torch.int64).to(device), attn_mask], dim=1
                )

            labels = encoded_batch

            with torch.no_grad():
                out_logits = model(encoded_batch, attention_mask=attn_mask).logits

            shift_logits = out_logits[..., :-1, :].contiguous()
            shift_labels = labels[..., 1:].contiguous()
            shift_attention_mask_batch = attn_mask[..., 1:].contiguous()

            perplexity_batch = torch.exp(
                (loss_fct(shift_logits.transpose(1, 2), shift_labels) * shift_attention_mask_batch).sum(1)
                / shift_attention_mask_batch.sum(1)
            )

            ppls += perplexity_batch.tolist()

        return {"perplexities": ppls, "mean_perplexity": np.mean(ppls)}