File size: 11,590 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": 86.4,
        "origin_prompt": "Examine the given sentence and decide if it is grammatically sound. Answer with either 'Acceptable' or 'Unacceptable'.",
        "atk_acc": 39.5,
        "atk_prompt": "Inspect the made condemning and select if it is grammatically healthier. Answers with either 'Acceptable' or 'Unacceptable'. ",
        "transfer_ori_acc": 81.0,
        "transfer_atk_acc": 77.0
    },
    {
        "dataset": "cola",
        "type": "role",
        "origin_acc": 86.6,
        "origin_prompt": "While performing grammar analysis, classify the grammar of the following sentence as 'acceptable' for correct grammar or 'unacceptable' for incorrect grammar.",
        "atk_acc": 73.0,
        "atk_prompt": "While performing grammar scans, classify the verbs of the afterward condemned as 'acceptable' for correctness grammar or 'unacceptable' for inappropriate vocabulary. ",
        "transfer_ori_acc": 81.5,
        "transfer_atk_acc": 80.5
    },
    {
        "dataset": "mrpc",
        "type": "task",
        "origin_acc": 86.52,
        "origin_prompt": "Compare the following sentences and determine if they share the same semantic meaning by responding with 'equivalent' or 'not_equivalent'.",
        "atk_acc": 83.58,
        "atk_prompt": "Compare the later judgements and defines if they exchange the same semantic sensation by answered with 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 73.5,
        "transfer_atk_acc": 73.0
    },
    {
        "dataset": "mrpc",
        "type": "role",
        "origin_acc": 89.22,
        "origin_prompt": "As a sentence similarity evaluator, analyze the provided sentences and indicate if their meanings are 'equivalent' or 'not_equivalent'.",
        "atk_acc": 85.29,
        "atk_prompt": "As a sentence metaphor evaluator, analyze the brought priors and demonstrating if their meanings are 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 73.5,
        "transfer_atk_acc": 65.5
    },
    {
        "dataset": "qnli",
        "type": "task",
        "origin_acc": 94.0,
        "origin_prompt": "Analyze the context and question, and indicate if the context entails the answer by choosing 'entailment' or 'not_entailment'.",
        "atk_acc": 54.2,
        "atk_prompt": "Explore the context and question, and suggest if the context involves the rebuttal by selected 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 80.5,
        "transfer_atk_acc": 73.5
    },
    {
        "dataset": "qnli",
        "type": "role",
        "origin_acc": 93.9,
        "origin_prompt": "As a semantic researcher, evaluate whether the provided context supports the answer to the question and choose 'entailment' or 'not_entailment'.",
        "atk_acc": 61.5,
        "atk_prompt": "Because a semantic inquisitors, appraise whether the provided context alleges the rebuttal to the issuing and selector 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 82.5,
        "transfer_atk_acc": 80.5
    },
    {
        "dataset": "qqp",
        "type": "task",
        "origin_acc": 88.7,
        "origin_prompt": "Examine the following expressions and tell me if they are alike in meaning by using 'equivalent' or 'not_equivalent'.",
        "atk_acc": 85.9,
        "atk_prompt": "Examine the afterward words and saying me if they are uniformly in connotation by using 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 82.0,
        "transfer_atk_acc": 81.5
    },
    {
        "dataset": "qqp",
        "type": "role",
        "origin_acc": 90.0,
        "origin_prompt": "As a tool for determining question equivalence, review the questions and categorize their similarity as either 'equivalent' or 'not_equivalent'.",
        "atk_acc": 87.6,
        "atk_prompt": "Because a tool for determining question equivalence, examination the queries and classifications their likeness as either 'equivalent' or 'not_equivalent'. ",
        "transfer_ori_acc": 83.5,
        "transfer_atk_acc": 83.0
    },
    {
        "dataset": "rte",
        "type": "task",
        "origin_acc": 93.5,
        "origin_prompt": "Identify whether the given pair of sentences demonstrates entailment or not_entailment. Answer with 'entailment' or 'not_entailment'.",
        "atk_acc": 92.42,
        "atk_prompt": "Deciding whether the given pair of sentencing demonstrates entailment or not_entailment. Answer with 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 82.0,
        "transfer_atk_acc": 82.5
    },
    {
        "dataset": "rte",
        "type": "role",
        "origin_acc": 93.5,
        "origin_prompt": "As a tool for determining entailment relationships, review the two statements and categorize their connection as either 'entailment' or 'not_entailment'.",
        "atk_acc": 91.34,
        "atk_prompt": "Because a device for determining entailment relationships, review the two words and categorize their connection as either 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 83.0,
        "transfer_atk_acc": 81.5
    },
    {
        "dataset": "sst2",
        "type": "task",
        "origin_acc": 96.79,
        "origin_prompt": "Determine the overall sentiment of this sentence, categorizing it as 'positive' or 'negative'.",
        "atk_acc": 93.46,
        "atk_prompt": "Determine the plenary senses of this condemning, categorizing it as 'positive' or 'negative'. ",
        "transfer_ori_acc": 95.0,
        "transfer_atk_acc": 95.0
    },
    {
        "dataset": "sst2",
        "type": "role",
        "origin_acc": 96.56,
        "origin_prompt": "As a sentiment classifier, determine whether the following text is 'positive' or 'negative'.",
        "atk_acc": 95.18,
        "atk_prompt": "As a emotion classifier, ascertain whether the suite wording is 'positive' or 'negative'. ",
        "transfer_ori_acc": 97.5,
        "transfer_atk_acc": 94.5
    },
    {
        "dataset": "wnli",
        "type": "task",
        "origin_acc": 77.46,
        "origin_prompt": "Assess the connection between the following sentences and classify it as 'entailment' or 'not_entailment'.",
        "atk_acc": 73.24,
        "atk_prompt": "Appraisal the attach between the suite condemns and classify it as 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 63.38028169014085,
        "transfer_atk_acc": 67.6056338028169
    },
    {
        "dataset": "wnli",
        "type": "role",
        "origin_acc": 76.06,
        "origin_prompt": "In your role as an entailment analysis tool, assess the relationship between the given sentences and classify it as 'entailment' or 'not_entailment'.",
        "atk_acc": 73.24,
        "atk_prompt": "In your role as an entailment analysis tool, assess the reports between the given sentences and classify it as 'entailment' or 'not_entailment'. ",
        "transfer_ori_acc": 63.38028169014085,
        "transfer_atk_acc": 61.97183098591549
    },
    {
        "dataset": "mmlu",
        "type": "task",
        "origin_acc": 54.08,
        "origin_prompt": "For the multiple-choice question related to {}, please choose the most accurate answer from 'A', 'B', 'C', or 'D'.",
        "atk_acc": 24.82,
        "atk_prompt": "Within the multiple-choice item relevant to {}, please decides the most careful rebuttal from 'A', 'B', 'C', or 'D'. ",
        "transfer_ori_acc": 66.5,
        "transfer_atk_acc": 63.0
    },
    {
        "dataset": "mmlu",
        "type": "role",
        "origin_acc": 53.72,
        "origin_prompt": "As an expert in {}, respond to the following multiple-choice question by selecting 'A', 'B', 'C', or 'D'.",
        "atk_acc": 51.95,
        "atk_prompt": "As an whiz in {}, respond to the following multiple-choice question by decided 'A', 'B', 'C', or 'D'.",
        "transfer_ori_acc": 65.5,
        "transfer_atk_acc": 64.0
    },
    {
        "dataset": "squad_v2",
        "type": "task",
        "origin_acc": 91.05,
        "origin_prompt": "Please derive the most fitting answer from the context. If there isn't an answer in the context, use 'unanswerable'.",
        "atk_acc": 55.32,
        "atk_prompt": "Ask subtract the most fitting rebuttal from the context. If there isn't an rebuttal in the context, use 'unanswerable'.",
        "transfer_ori_acc": 73.99378231946882,
        "transfer_atk_acc": 73.43394270771921
    },
    {
        "dataset": "squad_v2",
        "type": "role",
        "origin_acc": 90.55,
        "origin_prompt": "Drawing upon your expertise in the context, determine the most suitable answer. If an answer isn't available, state 'unanswerable'.",
        "atk_acc": 55.41,
        "atk_prompt": "Drawing upon your acquaintances in the context, identify the most suitable rebuttal. If an answer isn't available, state 'unanswerable'.",
        "transfer_ori_acc": 71.40309414291109,
        "transfer_atk_acc": 70.88917244194005
    },
    {
        "dataset": "iwslt",
        "type": "task",
        "origin_acc": 0.27,
        "origin_prompt": "Accurately translate the sentence from {} to {}, ensuring the meaning remains intact.",
        "atk_acc": 0.19,
        "atk_prompt": "Rightly translate the convicts from {} to {}, insurance the signify sojourn unaffected. ",
        "transfer_ori_acc": 0.2374651517338429,
        "transfer_atk_acc": 0.23893397344732503
    },
    {
        "dataset": "iwslt",
        "type": "role",
        "origin_acc": 0.27,
        "origin_prompt": "Assuming the responsibilities of a professional translator, translate the subsequent text passage from {} to {}.",
        "atk_acc": 0.19,
        "atk_prompt": "Adopt the indebted of a vocational artist, translate the subsequent laws transition from {} to {}. ",
        "transfer_ori_acc": 0.24394206967987916,
        "transfer_atk_acc": 0.250075978274994
    },
    {
        "dataset": "un_multi",
        "type": "task",
        "origin_acc": 0.36,
        "origin_prompt": "Perform a machine translation for the provided sentence, changing it from {} to {}.",
        "atk_acc": 0.35,
        "atk_prompt": "Perform a computer translate for the tabled sentence, changing it from {} to {}. ",
        "transfer_ori_acc": 0.352420428735174,
        "transfer_atk_acc": 0.35238062711589224
    },
    {
        "dataset": "un_multi",
        "type": "role",
        "origin_acc": 0.36,
        "origin_prompt": "As a skilled linguist, please convert the following sentence from {} to {}.",
        "atk_acc": 0.35,
        "atk_prompt": "As a skilled linguist, please translate the following punishment from {} to {}. ",
        "transfer_ori_acc": 0.3496473671163302,
        "transfer_atk_acc": 0.3501481660383654
    },
    {
        "dataset": "math",
        "type": "task",
        "origin_acc": 16.25,
        "origin_prompt": "Calculate the answer to the following math query about {}.",
        "atk_acc": 10.0,
        "atk_prompt": "Estimate the rebuttal to the following math view about {}. ",
        "transfer_ori_acc": 21.25,
        "transfer_atk_acc": 10.0
    },
    {
        "dataset": "math",
        "type": "role",
        "origin_acc": 16.25,
        "origin_prompt": "Act as a math expert and compute the result of the following mathematical task about {}.",
        "atk_acc": 13.75,
        "atk_prompt": "Act as a math expert and estimate the conclusion of the following mathematical task about {}. ",
        "transfer_ori_acc": 20.625,
        "transfer_atk_acc": 19.375
    }
]