File size: 12,074 Bytes
683d0f6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
[
    {
        "dataset": "cola",
        "type": "task",
        "origin_acc": 77.5,
        "origin_prompt": "Determine if the grammar of the given sentence is 'Acceptable' or 'Unacceptable':",
        "atk_acc": 31.0,
        "atk_prompt": "Establishes if the jargon of the made punishments is 'Acceptable' or 'Unacceptable':",
        "transfer_ori_acc": 86.0,
        "transfer_atk_acc": 41.099999999999994
    },
    {
        "dataset": "cola",
        "type": "role",
        "origin_acc": 76.3,
        "origin_prompt": "As an instrument for grammar evaluation, consider the sentence and determine if its grammar is correct, responding with 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar:",
        "atk_acc": 30.8,
        "atk_prompt": "Since an apparatus for language measured, envisage the convicts and identify if its language is correct, reaction with 'acceptable' for accurate jargon or 'unacceptable' for unsound pronunciation:",
        "transfer_ori_acc": 86.3,
        "transfer_atk_acc": 80.2
    },
    {
        "dataset": "mrpc",
        "type": "task",
        "origin_acc": 82.11,
        "origin_prompt": "Do these two sentences have the same underlying meaning? Respond with 'equivalent' or 'not_equivalent'.",
        "atk_acc": 53.68,
        "atk_prompt": "Ca these two condemning ont the same main meaning? Reactions with 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 86.02941176470588,
        "transfer_atk_acc": 83.57843137254902
    },
    {
        "dataset": "mrpc",
        "type": "role",
        "origin_acc": 81.62,
        "origin_prompt": "As a language comparison expert, examine the given pair of sentences and decide if their meanings align, answering with 'equivalent' or 'not_equivalent'.",
        "atk_acc": 12.01,
        "atk_prompt": "As a language compared professional, reviews the given torque of sentences and picked if their sens unify, address with 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 86.27450980392157,
        "transfer_atk_acc": 86.51960784313727
    },
    {
        "dataset": "qnli",
        "type": "task",
        "origin_acc": 89.8,
        "origin_prompt": "Evaluate whether the given context supports the answer to the question by responding with 'entailment' or 'not_entailment'.",
        "atk_acc": 41.4,
        "atk_prompt": "Estimation whether the gave backgrounds perpetuates the answered to the trouble by responses with 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 93.89999999999999,
        "transfer_atk_acc": 93.89999999999999
    },
    {
        "dataset": "qnli",
        "type": "role",
        "origin_acc": 89.4,
        "origin_prompt": "As a semantic interpreter, assess whether the provided context supports the answer to the given question and answer with 'entailment' or 'not_entailment'.",
        "atk_acc": 3.3,
        "atk_prompt": "Iike a semantic performers, valuing whether the afforded ambient perpetuates the answer to the rendered item and cope with 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 94.6,
        "transfer_atk_acc": 85.5
    },
    {
        "dataset": "qqp",
        "type": "task",
        "origin_acc": 87.7,
        "origin_prompt": "Assess whether the following statements are identical in meaning by answering 'equivalent' or 'not_equivalent'.",
        "atk_acc": 18.9,
        "atk_prompt": "Valuing whether the later languages are comparable in sinn by cope 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 87.7,
        "transfer_atk_acc": 87.7
    },
    {
        "dataset": "qqp",
        "type": "role",
        "origin_acc": 86.9,
        "origin_prompt": "In the role of a question similarity analyzer, examine the meaning of the provided questions and decide if they are equivalent, answering with 'equivalent' for equivalent questions or 'not_equivalent' for different questions.",
        "atk_acc": 73.8,
        "atk_prompt": "About the rol of a question likeness analyzer, examine the mean of the conferred topic and identifying if they are equivalent, retaliated with 'equivalent' for equivalent trouble or 'not_equivalent' for other topics. ",
        "transfer_ori_acc": 88.1,
        "transfer_atk_acc": 89.1
    },
    {
        "dataset": "rte",
        "type": "task",
        "origin_acc": 86.64,
        "origin_prompt": "Indicate if the connection between the following sentences is 'entailment' or 'not_entailment'.",
        "atk_acc": 80.87,
        "atk_prompt": "Portrays if the login between the following punishments is 'entailment' or 'not_entailment'.",
        "transfer_ori_acc": 92.7797833935018,
        "transfer_atk_acc": 92.4187725631769
    },
    {
        "dataset": "rte",
        "type": "role",
        "origin_acc": 84.84,
        "origin_prompt": "In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment' or 'not_entailment':",
        "atk_acc": 79.06,
        "atk_prompt": "About the capacity of an entailment weighed system, specified if the link between the following phrases is 'entailment' or 'not_entailment':",
        "transfer_ori_acc": 92.7797833935018,
        "transfer_atk_acc": 91.69675090252709
    },
    {
        "dataset": "sst2",
        "type": "task",
        "origin_acc": 94.95,
        "origin_prompt": "Given the context of this text, indicate if the emotion conveyed is 'positive' or 'negative':",
        "atk_acc": 84.29,
        "atk_prompt": "Made the background of this version, depicts if the passion recharged is 'positive' or 'negative': ",
        "transfer_ori_acc": 95.41284403669725,
        "transfer_atk_acc": 95.06880733944955
    },
    {
        "dataset": "sst2",
        "type": "role",
        "origin_acc": 95.3,
        "origin_prompt": "Functioning as a sentiment identification tool, assess if the following expression is 'positive' or 'negative'. Classify:",
        "atk_acc": 25.46,
        "atk_prompt": "Functions as a emotion defining instrumentation, valuing if the next voicing is 'positive' or 'negative'. Classification: ",
        "transfer_ori_acc": 95.9862385321101,
        "transfer_atk_acc": 94.83944954128441
    },
    {
        "dataset": "wnli",
        "type": "task",
        "origin_acc": 66.2,
        "origin_prompt": "Are the following two sentences entailment or not_entailment? Answer me with \"entailment\" or \"not_entailment\", just one word.",
        "atk_acc": 56.34,
        "atk_prompt": "Be the following two phrases entailment or not_entailment? Responded me with \"entailment\" or \"not_entailment\", just one verb. ",
        "transfer_ori_acc": 78.87323943661971,
        "transfer_atk_acc": 74.64788732394366
    },
    {
        "dataset": "wnli",
        "type": "role",
        "origin_acc": 64.79,
        "origin_prompt": "In the capacity of an entailment assessment system, indicate if the link between the following sentences is 'entailment' or 'not_entailment':",
        "atk_acc": 56.34,
        "atk_prompt": "Among the capacity of an entailment analysis system, indicate if the attach between the subsequently sentences is 'entailment' or 'not_entailment':",
        "transfer_ori_acc": 78.87323943661971,
        "transfer_atk_acc": 74.64788732394366
    },
    {
        "dataset": "mmlu",
        "type": "task",
        "origin_acc": 45.74,
        "origin_prompt": "For each multiple-choice question about {}, identify the correct answer by selecting 'A', 'B', 'C', or 'D'.",
        "atk_acc": 37.06,
        "atk_prompt": "By any multiple-choice question about {}, locate the corrected riposte by akhtar 'A', 'B', 'C', or 'D'. ",
        "transfer_ori_acc": 51.77304964539007,
        "transfer_atk_acc": 52.836879432624116
    },
    {
        "dataset": "mmlu",
        "type": "role",
        "origin_acc": 44.86,
        "origin_prompt": "Utilizing your expertise in {}, answer the following multiple-choice question by picking 'A', 'B', 'C', or 'D'.",
        "atk_acc": 40.96,
        "atk_prompt": "Utilise your expertise in {}, meet the next multiple-choice question by collection 'A', 'B', 'C', or 'D'.",
        "transfer_ori_acc": 52.12765957446809,
        "transfer_atk_acc": 53.54609929078015
    },
    {
        "dataset": "squad_v2",
        "type": "task",
        "origin_acc": 87.65,
        "origin_prompt": "Please derive the most fitting answer from the context. If there isn't an answer in the context, respond with 'unanswerable'.",
        "atk_acc": 84.09,
        "atk_prompt": "Please subtract the most adjust answered from the context. Until there isn't an replied in the context, answered with 'unanswerable'.",
        "transfer_ori_acc": 89.88229949874687,
        "transfer_atk_acc": 89.98229949874687
    },
    {
        "dataset": "squad_v2",
        "type": "role",
        "origin_acc": 88.05,
        "origin_prompt": "As an authority on the context, locate the most accurate answer. If the context doesn't contain the answer, mention 'unanswerable'.",
        "atk_acc": 77.31,
        "atk_prompt": "As an authority on the context, find the most proper rebuttal. Though the context doesn't incorporated the answering, speaks 'unanswerable'.",
        "transfer_ori_acc": 88.98646616541353,
        "transfer_atk_acc": 55.74122807017544
    },
    {
        "dataset": "iwslt",
        "type": "task",
        "origin_acc": 0.14,
        "origin_prompt": "Please perform a translation of the given sentence, converting it from {} to {}:",
        "atk_acc": 0.13,
        "atk_prompt": "Please accomplishing a translators of the given judging, converting it from {} to {}:",
        "transfer_ori_acc": 0.17287994792239936,
        "transfer_atk_acc": 0.17242775573031324
    },
    {
        "dataset": "iwslt",
        "type": "role",
        "origin_acc": 0.14,
        "origin_prompt": "In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}:",
        "atk_acc": 0.06,
        "atk_prompt": "Under the features of a parlance specializing, realization a typewriter translator for the envisaged chastisement, altering it from {} to {}: ",
        "transfer_ori_acc": 0.1757473985410512,
        "transfer_atk_acc": 0.14220973997645037
    },
    {
        "dataset": "un_multi",
        "type": "task",
        "origin_acc": 0.27,
        "origin_prompt": "Convert the following sentence from its original {} language to the target language {}:",
        "atk_acc": 0.11,
        "atk_prompt": "Transformations the later sanctions from its preliminary {} linguistics to the fated parlance {}:",
        "transfer_ori_acc": 0.31092395135800066,
        "transfer_atk_acc": 0.25185973744569756
    },
    {
        "dataset": "un_multi",
        "type": "role",
        "origin_acc": 0.27,
        "origin_prompt": "In the role of a language expert, perform a machine translation for the provided sentence, changing it from {} to {}:",
        "atk_acc": 0.15,
        "atk_prompt": "Under the role of a language skilled, performed a machine interpreters for the furnished condemned, transform it from {} to {}: ",
        "transfer_ori_acc": 0.31256243540663714,
        "transfer_atk_acc": 0.2918302355274189
    },
    {
        "dataset": "math",
        "type": "task",
        "origin_acc": 13.75,
        "origin_prompt": "Work out the answer to this math problem featuring {}:",
        "atk_acc": 1.88,
        "atk_prompt": "Work out the address to this math problem idiosyncratic {}:",
        "transfer_ori_acc": 12.5,
        "transfer_atk_acc": 4.375
    },
    {
        "dataset": "math",
        "type": "role",
        "origin_acc": 13.12,
        "origin_prompt": "As a mathematics instructor, calculate the answer to the following problem related to {}:",
        "atk_acc": 1.25,
        "atk_prompt": "As a mathematics prof, calculate the address to the following problem related to {}: ",
        "transfer_ori_acc": 15.0,
        "transfer_atk_acc": 4.375
    }
]