__init__.py 4.4 KB
Newer Older
&'s avatar
& committed
1
2
from pprint import pprint

&'s avatar
& committed
3
4
import sacrebleu

Jason Phang's avatar
Jason Phang committed
5
6
from . import superglue
from . import glue
Leo Gao's avatar
Leo Gao committed
7
from . import arc
thefazzer's avatar
thefazzer committed
8
from . import coqa
Leo Gao's avatar
Leo Gao committed
9
from . import race
Leo Gao's avatar
Leo Gao committed
10
from . import webqs
Leo Gao's avatar
Leo Gao committed
11
from . import anli
12
from . import wsc273
Charles Foster's avatar
Charles Foster committed
13
from . import winogrande
Charles Foster's avatar
Charles Foster committed
14
from . import quac
Charles Foster's avatar
Charles Foster committed
15
from . import hellaswag
Charles Foster's avatar
Charles Foster committed
16
from . import openbookqa
Charles Foster's avatar
Charles Foster committed
17
from . import squad
18
from . import naturalqs
19
from . import sat
20
from . import arithmetic
Leo Gao's avatar
Leo Gao committed
21
from . import lambada
Jon Tow's avatar
Jon Tow committed
22
from . import race
Leo Gao's avatar
Leo Gao committed
23
from . import piqa
24
from . import triviaqa
jeffhsu3's avatar
jeffhsu3 committed
25
from . import pubmedqa
26
from . import sciq
27
from . import webqs
28
from . import qa4mre
&'s avatar
& committed
29
from . import translation
30
31
from . import headqa
from . import mathqa
32
from . import unscramble
Jason Phang's avatar
gpt3  
Jason Phang committed
33

&'s avatar
& committed
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
########################################
# Translation tasks
########################################

# 6 total
gpt3_translation_benchmarks = {
    "wmt14": ['en-fr', 'fr-en'],  # French
    "wmt16": ['en-ro', 'ro-en', 'de-en', 'en-de'],  # German, Romanian
}


# 28 total
selected_translation_benchmarks = {
    **gpt3_translation_benchmarks,
    "wmt20": sacrebleu.get_langpairs_for_testset("wmt20"),
    "iwslt17": ['en-ar', 'ar-en']  # Arabic
}

# 319 total
all_translation_benchmarks = {
    ts: sacrebleu.get_langpairs_for_testset(ts)
    for ts in sacrebleu.get_available_testsets()
}


########################################
# All tasks
########################################


Jason Phang's avatar
Jason Phang committed
64
TASK_REGISTRY = {
Jason Phang's avatar
multirc  
Jason Phang committed
65
    # GLUE
Jason Phang's avatar
Jason Phang committed
66
67
    "cola": glue.CoLA,
    "mnli": glue.MNLI,
Jason Phang's avatar
Jason Phang committed
68
    "mnli_mismatched": glue.MNLIMismatched,
Jason Phang's avatar
Jason Phang committed
69
70
71
72
    "mrpc": glue.MRPC,
    "rte": glue.RTE,
    "qnli": glue.QNLI,
    "qqp": glue.QQP,
73
    #"stsb": glue.STSB, # not implemented yet
Jason Phang's avatar
Jason Phang committed
74
75
    "sst": glue.SST,
    "wnli": glue.WNLI,
Jason Phang's avatar
multirc  
Jason Phang committed
76
    # SuperGLUE
Jason Phang's avatar
Jason Phang committed
77
    "boolq": superglue.BoolQ,
thefazzer's avatar
thefazzer committed
78
    "cb": superglue.CommitmentBank,
Jason Phang's avatar
Jason Phang committed
79
    "copa": superglue.Copa,
Jason Phang's avatar
multirc  
Jason Phang committed
80
    "multirc": superglue.MultiRC,
Jason Phang's avatar
Jason Phang committed
81
    "record": superglue.ReCoRD,
Jason Phang's avatar
Jason Phang committed
82
    "wic": superglue.WordsInContext,
Jason Phang's avatar
wsc  
Jason Phang committed
83
    "wsc": superglue.SGWinogradSchemaChallenge,
84
    
Jason Phang's avatar
multirc  
Jason Phang committed
85
    # Order by benchmark/genre?
thefazzer's avatar
thefazzer committed
86
    "coqa": coqa.CoQA,
Leo Gao's avatar
Leo Gao committed
87
    "lambada": lambada.LAMBADA,
Leo Gao's avatar
Leo Gao committed
88
    "piqa": piqa.PiQA,
Leo Gao's avatar
Leo Gao committed
89

90
    # Science related
jeffhsu3's avatar
jeffhsu3 committed
91
    "pubmedqa" : pubmedqa.Pubmed_QA,
92
    "sciq" : sciq.SciQ,
93
94
95
96
    #"qa4mre" : qa4mre.QA4MRE,
    "qa4mre_2011" : qa4mre.QA4MRE_2011,
    "qa4mre_2012" : qa4mre.QA4MRE_2012,
    "qa4mre_2013" : qa4mre.QA4MRE_2013,
jeffhsu3's avatar
jeffhsu3 committed
97

Leo Gao's avatar
Leo Gao committed
98
    #"triviaqa": triviaqa.TriviaQA,
Jonathan Tow's avatar
Jonathan Tow committed
99
100
    "arc_easy": arc.ARCEasy,
    "arc_challenge": arc.ARCChallenge,
101
    # "quac": quac.QuAC, # not implemented yet
102
    "hellaswag": hellaswag.HellaSwag, # not implemented yet
Jon Tow's avatar
Jon Tow committed
103
    "openbookqa": openbookqa.OpenBookQA,
104
105
    # "sat": sat.SATAnalogies, # not implemented yet
    # "squad": squad.SQuAD, # not implemented yet
Jon Tow's avatar
Jon Tow committed
106
    "race": race.RACE,
107
    # "naturalqs": naturalqs.NaturalQs, # not implemented yet
108
109
    "headqa": headqa.HeadQA,
    "mathqa": mathqa.MathQA,
110
    "webqs": webqs.WebQs,
111
    "wsc273": wsc273.WinogradSchemaChallenge273,
112
    "winogrande": winogrande.Winogrande,
Jonathan Tow's avatar
Jonathan Tow committed
113
114
115
    "anli_r1": anli.ANLIRound1,
    "anli_r2": anli.ANLIRound2,
    "anli_r3": anli.ANLIRound3,
116
117
118
119
120
121
122
123
124
125
126
    # arithmetic
    "arithmetic_2da": arithmetic.Arithmetic2DPlus,
    "arithmetic_2ds": arithmetic.Arithmetic2DMinus,
    "arithmetic_3da": arithmetic.Arithmetic3DPlus,
    "arithmetic_3ds": arithmetic.Arithmetic3DMinus,
    "arithmetic_4da": arithmetic.Arithmetic4DPlus,
    "arithmetic_4ds": arithmetic.Arithmetic4DMinus,
    "arithmetic_5da": arithmetic.Arithmetic5DPlus,
    "arithmetic_5ds": arithmetic.Arithmetic5DMinus,
    "arithmetic_2dm": arithmetic.Arithmetic2DMultiplication,
    "arithmetic_1dc": arithmetic.Arithmetic1DComposite,
&'s avatar
& committed
127
128
    # TODO Perhaps make these groups of tasks
    #   e.g. anli, arithmetic, openai_translations, harness_translations
&'s avatar
...  
& committed
129
130

    # e.g. wmt14-fr-en
&'s avatar
& committed
131
132
133
    **translation.create_tasks_from_benchmarks(gpt3_translation_benchmarks),
    # chef's selection, mostly wmt20
    **translation.create_tasks_from_benchmarks(selected_translation_benchmarks),
Jon Tow's avatar
Jon Tow committed
134
135

    # Word Scrambling and Manipulation Tasks
136
137
138
139
140
    "anagrams1": unscramble.Anagrams1,
    "anagrams2": unscramble.Anagrams2,
    "cycle_letters": unscramble.CycleLetters,
    "random_insertion": unscramble.RandomInsertion,
    "reversed_words": unscramble.ReversedWords,
Jason Phang's avatar
Jason Phang committed
141
}
Jason Phang's avatar
gpt3  
Jason Phang committed
142
143


Jason Phang's avatar
Jason Phang committed
144
ALL_TASKS = sorted(list(TASK_REGISTRY))
Jason Phang's avatar
Jason Phang committed
145
146


Jason Phang's avatar
cleanup  
Jason Phang committed
147
def get_task(task_name):
&'s avatar
& committed
148
149
150
151
152
153
    try:
        return TASK_REGISTRY[task_name]
    except KeyError as e:
        print("Available tasks:")
        pprint(TASK_REGISTRY)
        raise KeyError(f"Missing task {task_name}")
Jason Phang's avatar
cleanup  
Jason Phang committed
154
155
156
157
158
159
160


def get_task_dict(task_name_list):
    return {
        task_name: get_task(task_name)()
        for task_name in task_name_list
    }