File size: 11,970 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": 77.9,
        "origin_prompt": "Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable':",
        "atk_acc": 30.9,
        "atk_prompt": "examine the english plural and say if it is grammaticalmost sound. Answer with neither 'Acceptable' or 'Unacceptable':",
        "transfer_ori_acc": 86.1,
        "transfer_atk_acc": 77.60000000000001
    },
    {
        "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": 30.9,
        "atk_prompt": "for an organ for linguistic evaluation, write the movie and question if its grammar is incorrect, and with 'acceptable' for new grammar or 'unacceptable' for unstable grammar:",
        "transfer_ori_acc": 86.3,
        "transfer_atk_acc": 83.6
    },
    {
        "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": 16.67,
        "atk_prompt": "do these two clauses have the same underlying meanings? deal with 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 86.02941176470588,
        "transfer_atk_acc": 84.80392156862744
    },
    {
        "dataset": "mrpc",
        "type": "role",
        "origin_acc": 82.6,
        "origin_prompt": "In your capacity as a semantic comparison tool, examine the given sentences and decide if their meanings are identical, responding with 'equivalent' or 'not_equivalent'.",
        "atk_acc": 9.07,
        "atk_prompt": "using your capacity as a semantic relations member, determine the given words and decide if their meanings are consecutive, comparison with 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 87.00980392156863,
        "transfer_atk_acc": 82.35294117647058
    },
    {
        "dataset": "qnli",
        "type": "task",
        "origin_acc": 88.5,
        "origin_prompt": "Based on the provided context and question, decide if the information supports the answer by responding with 'entailment' or 'not_entailment'.",
        "atk_acc": 27.6,
        "atk_prompt": "following on the supplied context and questioning, save if the information states the response by simply with 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 94.8,
        "transfer_atk_acc": 61.3
    },
    {
        "dataset": "qnli",
        "type": "role",
        "origin_acc": 89.2,
        "origin_prompt": "As a textual analyst, examine if the given context logically implies the answer to the question and indicate your decision with 'entailment' or 'not_entailment'.",
        "atk_acc": 11.8,
        "atk_prompt": "as a textual analyst, scrutiny if the following means logical) implies the reply to the answer and cross your r with 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 93.5,
        "transfer_atk_acc": 93.8
    },
    {
        "dataset": "qqp",
        "type": "task",
        "origin_acc": 88.1,
        "origin_prompt": "Are the following two questions equivalent or not? Answer me with \"equivalent\" or \"not_equivalent\".",
        "atk_acc": 80.2,
        "atk_prompt": "to the next enough being equivalent or not? Answer me with \"equivalent\" or \"not_equivalent\". ",
        "transfer_ori_acc": 87.8,
        "transfer_atk_acc": 87.3
    },
    {
        "dataset": "qqp",
        "type": "role",
        "origin_acc": 87.7,
        "origin_prompt": "Acting as a question equivalence instrument, determine if the provided questions are equivalent in meaning, answering with 'equivalent' for similar questions or 'not_equivalent' for dissimilar ones.",
        "atk_acc": 72.2,
        "atk_prompt": "t as a question equal instrument, determine if the program questions are equivalent in semantic, asking with 'equivalent' for analogous question or 'not_equivalent' for nonssimicomplex times. ",
        "transfer_ori_acc": 87.9,
        "transfer_atk_acc": 87.4
    },
    {
        "dataset": "rte",
        "type": "task",
        "origin_acc": 86.64,
        "origin_prompt": "Are the following two sentences entailment or not_entailment? Answer me with \"entailment\" or \"not_entailment\", just one word.",
        "atk_acc": 81.59,
        "atk_prompt": "be the following deux verbs entailment or not_entailment? asking me with \"entailment\" or \"not_entailment\", just one word. ",
        "transfer_ori_acc": 94.22382671480143,
        "transfer_atk_acc": 92.05776173285199
    },
    {
        "dataset": "rte",
        "type": "role",
        "origin_acc": 84.84,
        "origin_prompt": "While performing entailment analysis, classify the relationship between the provided sentences as 'entailment' or 'not_entailment':",
        "atk_acc": 78.7,
        "atk_prompt": "While performing entailment analysis, construct the relationship between the and sentence as 'entailment' or 'not_entailment':",
        "transfer_ori_acc": 93.50180505415162,
        "transfer_atk_acc": 92.4187725631769
    },
    {
        "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": 92.89,
        "atk_prompt": "give the context of this context, and if the emotion recorded is 'positive' or 'negative': ",
        "transfer_ori_acc": 95.41284403669725,
        "transfer_atk_acc": 94.4954128440367
    },
    {
        "dataset": "sst2",
        "type": "role",
        "origin_acc": 95.41,
        "origin_prompt": "In the role of a sentiment analysis tool, respond with 'positive' or 'negative' to classify this statement:",
        "atk_acc": 89.79,
        "atk_prompt": "how the role of a compliment analysis tool, responses with 'positive' or 'negative' to mood this statement: ",
        "transfer_ori_acc": 96.44495412844036,
        "transfer_atk_acc": 96.44495412844036
    },
    {
        "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": 54.93,
        "atk_prompt": "are the following analogous sentence entailment or not_entailment? Answer me with \"entailment\" or \"not_entailment\", just one verb. ",
        "transfer_ori_acc": 78.87323943661971,
        "transfer_atk_acc": 73.23943661971832
    },
    {
        "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": "h the capacity of an entailment assessment system, indicate if the link between the following verb is 'entailment' or 'not_entailment':",
        "transfer_ori_acc": 78.87323943661971,
        "transfer_atk_acc": 64.7887323943662
    },
    {
        "dataset": "mmlu",
        "type": "task",
        "origin_acc": 45.57,
        "origin_prompt": "For the multiple-choice question related to {}, please choose the most accurate answer from 'A', 'B', 'C', or 'D'.",
        "atk_acc": 16.84,
        "atk_prompt": "For the multiple-choice choice similar to {}, please find the most inaccurate questions from 'A', 'B', 'C', or 'D'. ",
        "transfer_ori_acc": 53.36879432624113,
        "transfer_atk_acc": 13.829787234042554
    },
    {
        "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": 3.19,
        "atk_prompt": "Utilizing your expertise in {}, write the following multiple-choices question by each 'A', 'B', 'C', or 'D'.",
        "transfer_ori_acc": 52.12765957446809,
        "transfer_atk_acc": 40.95744680851064
    },
    {
        "dataset": "squad_v2",
        "type": "task",
        "origin_acc": 87.82,
        "origin_prompt": "Search the context for the most relevant answer. If the answer cannot be found, respond with 'unanswerable'.",
        "atk_acc": 36.51,
        "atk_prompt": "return the topic for the most mentioned question. then the answer cannot be found, respond with 'unanswerable'.",
        "transfer_ori_acc": 90.04896616541353,
        "transfer_atk_acc": 57.76345029239766
    },
    {
        "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": 42.92,
        "atk_prompt": "in an authority on the context, discover the most incorrect explanation. ensure the person doesn't contains the answer, mention 'unanswerable'.",
        "transfer_ori_acc": 88.98646616541353,
        "transfer_atk_acc": 44.5
    },
    {
        "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 perform a translation of the given citation, converting it from {} to {}:",
        "transfer_ori_acc": 0.17287994792239936,
        "transfer_atk_acc": 0.17666372994164403
    },
    {
        "dataset": "iwslt",
        "type": "role",
        "origin_acc": 0.14,
        "origin_prompt": "In the capacity of a proficient translator, please change the following sentence from {} to {}:",
        "atk_acc": 0.11,
        "atk_prompt": "i the capacity of a possessed translator, may modification the following sentence from {} to {}: ",
        "transfer_ori_acc": 0.17223574608131062,
        "transfer_atk_acc": 0.1709026262221616
    },
    {
        "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": 0.21,
        "atk_prompt": "refer the following sentence from its original {} language to the targets i {}:",
        "transfer_ori_acc": 0.31092395135800066,
        "transfer_atk_acc": 0.3089286594500741
    },
    {
        "dataset": "un_multi",
        "type": "role",
        "origin_acc": 0.27,
        "origin_prompt": "As a qualified translator, convert the given sentence from its original {} language to the target language {}:",
        "atk_acc": 0.17,
        "atk_prompt": "while a reflex translator, reverse the given sentence from its original {} language to the target languages {}: ",
        "transfer_ori_acc": 0.3122539424318205,
        "transfer_atk_acc": 0.30521816181759504
    },
    {
        "dataset": "math",
        "type": "task",
        "origin_acc": 14.37,
        "origin_prompt": "Obtain the result for this math question regarding {}:",
        "atk_acc": 2.5,
        "atk_prompt": "Obtain the procedure for this math question regarding {}:",
        "transfer_ori_acc": 13.750000000000002,
        "transfer_atk_acc": 0.0
    },
    {
        "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": 1.88,
        "atk_prompt": "As a mathematics instructor, calculate the sum to the following problem related to {}: ",
        "transfer_ori_acc": 15.0,
        "transfer_atk_acc": 8.75
    }
]