__init__.py 4.44 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
Jon Tow's avatar
Jon Tow committed
32
from . import drop
33
from . import unscramble
Jason Phang's avatar
gpt3  
Jason Phang committed
34

&'s avatar
& committed
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
########################################
# 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
65
TASK_REGISTRY = {
Jason Phang's avatar
multirc  
Jason Phang committed
66
    # GLUE
Jason Phang's avatar
Jason Phang committed
67
68
    "cola": glue.CoLA,
    "mnli": glue.MNLI,
Jason Phang's avatar
Jason Phang committed
69
    "mnli_mismatched": glue.MNLIMismatched,
Jason Phang's avatar
Jason Phang committed
70
71
72
73
    "mrpc": glue.MRPC,
    "rte": glue.RTE,
    "qnli": glue.QNLI,
    "qqp": glue.QQP,
74
    #"stsb": glue.STSB, # not implemented yet
Jason Phang's avatar
Jason Phang committed
75
76
    "sst": glue.SST,
    "wnli": glue.WNLI,
Jason Phang's avatar
multirc  
Jason Phang committed
77
    # SuperGLUE
Jason Phang's avatar
Jason Phang committed
78
    "boolq": superglue.BoolQ,
thefazzer's avatar
thefazzer committed
79
    "cb": superglue.CommitmentBank,
Jason Phang's avatar
Jason Phang committed
80
    "copa": superglue.Copa,
Jason Phang's avatar
multirc  
Jason Phang committed
81
    "multirc": superglue.MultiRC,
Jason Phang's avatar
Jason Phang committed
82
    "record": superglue.ReCoRD,
Jason Phang's avatar
Jason Phang committed
83
    "wic": superglue.WordsInContext,
Jason Phang's avatar
wsc  
Jason Phang committed
84
    "wsc": superglue.SGWinogradSchemaChallenge,
85
    
Jason Phang's avatar
multirc  
Jason Phang committed
86
    # Order by benchmark/genre?
thefazzer's avatar
thefazzer committed
87
    "coqa": coqa.CoQA,
Jon Tow's avatar
Jon Tow committed
88
    "drop": drop.DROP,
Leo Gao's avatar
Leo Gao committed
89
    "lambada": lambada.LAMBADA,
Leo Gao's avatar
Leo Gao committed
90
    "piqa": piqa.PiQA,
Leo Gao's avatar
Leo Gao committed
91

92
    # Science related
jeffhsu3's avatar
jeffhsu3 committed
93
    "pubmedqa" : pubmedqa.Pubmed_QA,
94
    "sciq" : sciq.SciQ,
95
96
97
98
    #"qa4mre" : qa4mre.QA4MRE,
    "qa4mre_2011" : qa4mre.QA4MRE_2011,
    "qa4mre_2012" : qa4mre.QA4MRE_2012,
    "qa4mre_2013" : qa4mre.QA4MRE_2013,
jeffhsu3's avatar
jeffhsu3 committed
99

Leo Gao's avatar
Leo Gao committed
100
    #"triviaqa": triviaqa.TriviaQA,
Jonathan Tow's avatar
Jonathan Tow committed
101
102
    "arc_easy": arc.ARCEasy,
    "arc_challenge": arc.ARCChallenge,
103
    # "quac": quac.QuAC, # not implemented yet
104
    "hellaswag": hellaswag.HellaSwag, # not implemented yet
Jon Tow's avatar
Jon Tow committed
105
    "openbookqa": openbookqa.OpenBookQA,
106
107
    # "sat": sat.SATAnalogies, # not implemented yet
    # "squad": squad.SQuAD, # not implemented yet
Jon Tow's avatar
Jon Tow committed
108
    "race": race.RACE,
109
    # "naturalqs": naturalqs.NaturalQs, # not implemented yet
110
111
    "headqa": headqa.HeadQA,
    "mathqa": mathqa.MathQA,
112
    "webqs": webqs.WebQs,
113
    "wsc273": wsc273.WinogradSchemaChallenge273,
114
    "winogrande": winogrande.Winogrande,
Jonathan Tow's avatar
Jonathan Tow committed
115
116
117
    "anli_r1": anli.ANLIRound1,
    "anli_r2": anli.ANLIRound2,
    "anli_r3": anli.ANLIRound3,
118
119
120
121
122
123
124
125
126
127
128
    # 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
129
130
    # TODO Perhaps make these groups of tasks
    #   e.g. anli, arithmetic, openai_translations, harness_translations
&'s avatar
...  
& committed
131
132

    # e.g. wmt14-fr-en
&'s avatar
& committed
133
134
135
    **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
136
137

    # Word Scrambling and Manipulation Tasks
138
139
140
141
142
    "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
143
}
Jason Phang's avatar
gpt3  
Jason Phang committed
144
145


Jason Phang's avatar
Jason Phang committed
146
ALL_TASKS = sorted(list(TASK_REGISTRY))
Jason Phang's avatar
Jason Phang committed
147
148


Jason Phang's avatar
cleanup  
Jason Phang committed
149
def get_task(task_name):
&'s avatar
& committed
150
151
152
153
154
155
    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
156
157
158
159
160
161
162


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