PromptBreeder/main_gpt.py at main · nahidabdollahi1998/PromptBreeder · GitHub

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
import os
import pandas as pd
import random
import re

import LLMs
import Evolution
import Mutation

# parameters
llm_model = 'gpt-3.5'
dataset_name = 'gsm8k'
problem_description = "Solve the math word problem, giving your answer as an arabic numeral."
train_test_size = 10 # it was 100 in Prompt Breeder
population_size = 50
epochs = 0


# load DATASET
train_path = llm_model + "_on_" + dataset_name + "/train.csv"
if os.path.exists(train_path) :
    df_train = pd.read_csv(train_path)
    train_list = df_train.to_numpy()
    df_test = pd.read_csv(llm_model + "_on_" + dataset_name + "/test.csv")
else :
    from datasets import load_dataset
    dataset = load_dataset(dataset_name, 'main')
    train_data = dataset['train']

    r = random.randrange(len(train_data) - (2 * train_test_size))
    train_list = []
    for i in range(r, (r + train_test_size)) :
        matches = re.findall(r'\b\d+\b', train_data['answer'][i])
        t_answer = int(matches[-1]) if matches else None
        train_list.append([train_data['question'][i], t_answer])
        df_train = pd.DataFrame(train_list)
        df_train.columns = ['question','answer']
        df_train.to_csv(train_path, index= False)
    test_list = []
    for i in range((r + train_test_size), (r + (2 * train_test_size))) :
        matches = re.findall(r'\b\d+\b', train_data['answer'][i])
        t_answer = int(matches[-1]) if matches else None
        test_list.append([train_data['question'][i], t_answer])
        df_test = pd.DataFrame(test_list)
        df_test.columns = ['question','answer']
        df_test.to_csv(llm_model + "_on_" + dataset_name + "/test.csv", index= False)

# initialize population
file_path = llm_model + "_on_" + dataset_name + "/population_initialize.csv"
if os.path.exists(file_path) :
    df_population = pd.read_csv(file_path)
    population = df_population.to_numpy()
else :
    population = Evolution.Evolution.Initialization(problem_description, population_size, llm_model,dataset_name)

print(len(population))

# Evaluate fitness
population = Evolution.Evolution.Evaluate_fitness(population[:20], train_list, llm_model,dataset_name, 0)

###### ITTERATION LOOP
for e in range(epochs) :
    ######  DO MUTATION
    mutated_population = []
    switch = random.randint(0,9)
    if switch == 0 : # ZERO ORDER PROMPT GENERATION
        Mutation.Direct_mutation.Zero_order_Prompt_Generation(population,llm_model,problem_description) #AMBIGUOUS

    if switch == 1 : # FIRST ORDER PROMPT GENERATION
        for i in range(population_size/2):
            if population[i][4] > population[i + (population_size/2)][4] : # compare fitness and choose the one with higher fitness
                p = population[i]
            else :
                p = population[i + (population_size/2)]
            m_1 = Mutation.Direct_mutation.First_order_Prompt_Generation(llm_model, p[1], p[0])
            m_2 = Mutation.Direct_mutation.First_order_Prompt_Generation(llm_model, p[3], p[2])
            # Append the choosen task-prompt and its mutated-task-prompt
            mutated_population.append([p[0],p[1],p[2],p[3]], [m_1,p[1],m_2,p[3]])

    if switch == 2 : # ESTIMATION OF DISTRIBUTION (EDA) MUTATION
        filtered_list, len_f_p = Mutation.Estimation_of_Distribution_mutation.Filtered_List(population)
        for i in range(population_size/2):
            # compare fitness and choose the one with higher fitness
            if population[i][4] > population[i + (population_size/2)][4] :
                p = population[i]
            else :
                p = population[i + (population_size/2)]
            m_1 = Mutation.Estimation_of_Distribution_mutation.EDA_Mutation(llm_model, filtered_list, p[1], e)
            m_2 = Mutation.Estimation_of_Distribution_mutation.EDA_Mutation(llm_model, filtered_list, p[3], e)
            # Append the choosen task-prompt and its mutated-task-prompt
            mutated_population.append([p[0],p[1],p[2],p[3]], [m_1,p[1],m_2,p[3]])

    if switch == 3 : # EDA RANK AND INDEX MUTATION
        filtered_list, len_f_p = Mutation.Estimation_of_Distribution_mutation.Filtered_List(population, True)
        for i in range(population_size/2):
            # compare fitness and choose the one with higher fitness
            if population[i][4] > population[i + (population_size/2)][4] :
                p = population[i]
            else :
                p = population[i + (population_size/2)]
            m_1 = Mutation.Estimation_of_Distribution_mutation.EDA_Rank_Index_Mutation(llm_model, filtered_list, len_f_p, p[1], e)
            m_2 = Mutation.Estimation_of_Distribution_mutation.EDA_Rank_Index_Mutation(llm_model, filtered_list, len_f_p, p[3], e)
            # Append the choosen task-prompt and its mutated-task-prompt
            mutated_population.append([p[0],p[1],p[2],p[3]], [m_1,p[1],m_2,p[3]])

    if switch == 4 : # LINEAGE BASED MUTATION
        list_best_so_far = Mutation.Estimation_of_Distribution_mutation.List_Best_So_Far(best_so_fars)
        for i in range(population_size/2):
            # compare fitness and choose the one with higher fitness
            if population[i][4] > population_[i + (population_size/2)][4] :
                p = population[i]
            else :
                p = population[i + (population_size/2)]
            m_1 = Mutation.Estimation_of_Distribution_mutation.Lineage_Based_Mutation(llm_model, list_best_so_far, p[1], e)
            m_2 = Mutation.Estimation_of_Distribution_mutation.Lineage_Based_Mutation(llm_model, list_best_so_far, p[3], e)
            # Append the choosen task-prompt and its mutated-task-prompt
            mutated_population.append([p[0],p[1],p[2],p[3]], [m_1,p[1],m_2,p[3]])

    if switch == 5 : # ZERO-ORDER HYPER-MUTATION
        df_t = pd.read_csv('thinking_styles.csv')
        thinking_styles = df_t.to_numpy().flatten()
        for i in range(population_size/2):
            # compare fitness and choose the one with higher fitness
            if population[i][4] > population[i + (population_size/2)][4] :
                p = population[i]
            else :
                p = population[i + (population_size/2)]
            t_1, m_1= Mutation.HyperMutation.Zero_Order_Hyper_Mutation(
                llm_model=llm_model,
                problem_description=problem_description,
                thinking_style=thinking_styles[random.randint(len(thinking_styles))],
                task_prompt=p[0],
                epoch=e)
            t_2, m_2 = Mutation.HyperMutation.Zero_Order_Hyper_Mutation(
                llm_model=llm_model,
                problem_description=problem_description,
                thinking_style=thinking_styles[random.randint(len(thinking_styles))],
                task_prompt=p[2],
                epoch=e)
            # Append the choosen task-prompt and its mutated-task-prompt
            mutated_population.append([p[0],p[1],p[2],p[3]], [t_1,m_1,t_2,m_2])

    if switch == 6 : # FIRST-ORDER HYPER-MUTATION
        for i in range(population_size/2):
            # compare fitness and choose the one with higher fitness
            if population[i][4] > population[i + (population_size/2)][4] :
                p = population[i]
            else :
                p = population[i + (population_size/2)]
            t_1, m_1 = Mutation.HyperMutation.First_Order_Hyper_Mutation(llm_model, p[0], p[1], e)
            t_2, m_2 = Mutation.HyperMutation.First_Order_Hyper_Mutation(llm_model, p[2], p[3], e)
            # Append the choosen task-prompt and its mutated-task-prompt
            mutated_population.append([p[0],p[1],p[2],p[3]], [t_1,m_1,t_2,m_2])

    if switch == 7 : # LAMARCKIAN MUTATION
        for i in range(population_size/2):
            # compare fitness and choose the one with higher fitness
            if population[i][4] > population[i + (population_size/2)][4] :
                p = population[i]
            else :
                p = population[i + (population_size/2)]
            r_workingout = random.randint(0,(len(train_list)-1))
            t_1 = Mutation.Lamarckian_mutation.Working_out_to_task_prompt(llm_model, train_list[r_workingout], train_list[r_workingout + 1], e)
            r_workingout = random.randint(0,(len(train_list)-1))
            t_2 = Mutation.Lamarckian_mutation.Working_out_to_task_prompt(llm_model, train_list[r_workingout], train_list[r_workingout + 1], e)
            # Append the choosen task-prompt and its mutated-task-prompt
            mutated_population.append([p[0],p[1],p[2],p[3]], [t_1,p[1],t_2,p[3]])

    if switch == 8 : # PROMPT CROSS-OVER
        mutated_population = Mutation.P_and_C.Prompt_Crossover(llm_model, population, e)

    #if switch == 9 : # CONTEXT SHUFFLING

    ###### EVALUATE NEW POPULATION
    population = Evolution.Evolution.Evaluate_fitness(mutated_population, train_list, llm_model, dataset_name, e + 1)