Recursive-Reasoning/generate_Q_tilde.py at main · SandipanMajhi/Recursive-Reasoning · 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
import torch
from argparse import ArgumentParser
from tqdm import tqdm
import json
import jsonlines
from generator import Gen_Model
from prompt import Promptconfig
from dataclasses import dataclass
from generation_prefix import retrieve_prompt_prefix
from generator import Gen_Model

@dataclass
class Tilde_Config:
    max_length: int = 64
    do_sample : bool = True
    temperature: float = 0.5
    top_p: float = 1.0
    return_dict_in_generate : bool = True
    output_scores : bool = True
    num_return_sequences: int = 1

@dataclass
class GlobalConfig:
    expt_name : str = "Tilde production"
    seed : int = -10
    dataset_name : str = "com2sense"

class Generate_Negation:
    def __init__(self, args):
        self.args = args
        self.general_prefix = open("promptsv_0_1/strategyqa/q_to_sent.txt", "r").read()
        self.prompt_prefix_dict = retrieve_prompt_prefix(args.dataset_name)
        self.prompt_configs = Promptconfig(self.prompt_prefix_dict["abductive"], self.prompt_prefix_dict["belief"],
                                           self.prompt_prefix_dict["negation"], self.prompt_prefix_dict["question"],
                                           self.prompt_prefix_dict["nli"])
        self.gen_model = Gen_Model("google/flan-t5-xl", seed=args.seed)
        self.device = "cuda" if torch.cuda.is_available() else "cpu"

    def prompt_sent(self,Q):
        generation_config = Tilde_Config()
        generation_config.temperature = self.args.temp
        prompt_str = self.create_general_sentence(Q)
        # print(f"Prompt str = {prompt_str}")
        num_Es_to_generate = generation_config.num_return_sequences
        E_list = []
        while len(E_list) < num_Es_to_generate:
            generation_config.num_return_sequences = num_Es_to_generate - len(E_list)
            response_inputs = self.gen_model.tokenizer(prompt_str, return_tensors = "pt").to(self.device)
            response = self.gen_model.model.generate(**response_inputs, **generation_config.__dict__)
            response = self.gen_model.tokenizer.batch_decode(response.sequences, skip_special_tokens = True)
            E_list.extend(Generate_Negation.filter_generated_explanations(response))
        # print(f"E_Tilde = {E_list[0]}")
        return E_list[0]

    def prompt_tilde(self,Q):
        generation_config = Tilde_Config()
        generation_config.temperature = self.args.temp
        prompt_str = self.prompt_configs.create_new_negation_prompt(Q)
        # print(f"prompt str = {prompt_str}")
        num_Es_to_generate = generation_config.num_return_sequences
        E_list = []
        while len(E_list) < num_Es_to_generate:
            generation_config.num_return_sequences = num_Es_to_generate - len(E_list)
            response_inputs = self.gen_model.tokenizer(prompt_str, return_tensors = "pt").to(self.device)
            response = self.gen_model.model.generate(**response_inputs, **generation_config.__dict__)
            response = self.gen_model.tokenizer.batch_decode(response.sequences, skip_special_tokens = True)
            E_list.extend(Generate_Negation.filter_generated_explanations(response))
        # print(f"E_Tilde = {E_list[0]}")
        return E_list[0]

    @staticmethod
    def filter_generated_explanations(explanations):
        filtered_explanations = [explanation.strip() for explanation in explanations]
        filtered_explanations = list(filter(lambda exp: len(exp) > 0 and exp.endswith("."), filtered_explanations))
        filtered_explanations = [explanation[0].upper() + explanation[1:] for explanation in filtered_explanations]
        if len(filtered_explanations) == 0:
            filtered_explanations.append(explanations[0].strip())
        filtered_explanations = list(dict.fromkeys(filtered_explanations))

        return filtered_explanations

    def create_general_sentence(self, Q):
        return f"{self.general_prefix}" \
            f"Q: {Q}\n"\
                f"A:"


if __name__ == "__main__":
    args = ArgumentParser()
    args.add_argument("--dataset_name", default = "com2sense", type = str)
    args.add_argument("--seed", default= 42, help=" Generation seed", type= int)
    args.add_argument("--temp", default = 0.5, type = float, help = "Generation temperature")

    args = args.parse_args()
    args.mode = "normal"

    if args.dataset_name == "com2sense":
        if args.mode == "normal":
            args.data_filename =  f"./data/{args.dataset_name}/dev.json"
            args.out_filename = f"./data/{args.dataset_name}/dev_Q_Q_tilde_{args.seed}"
    elif args.dataset_name == "csqa2":
        if args.mode == "normal":
            args.data_filename = f"./data/{args.dataset_name}/CSQA2_dev.json"
            args.out_filename = f"./data/{args.dataset_name}/dev_Q_Q_tilde_{args.seed}"
    elif args.dataset_name == "creak":
        if args.mode == "normal":
            args.data_filename = f"./data/{args.dataset_name}/dev.json"
            args.out_filename = f"./data/{args.dataset_name}/dev_Q_Q_tilde_{args.seed}"
    elif args.dataset_name == "strategyqa":
        print("Strategy QA does not have tilde mode")
        if args.mode == "normal":
            args.data_filename = f"./data/{args.dataset_name}/dev.json"
            args.out_filename = f"./data/{args.dataset_name}/dev_Q_Q_tilde_{args.seed}.json"


    if args.mode == "normal":
        if args.dataset_name == "csqa2":
            with jsonlines.open(args.data_filename, "r") as fp:
                samples = list(fp)
        elif args.dataset_name == "com2sense":
            with open(args.data_filename) as fp:
                samples = json.load(fp)
        elif args.dataset_name == "creak":
            with jsonlines.open(args.data_filename, "r") as fp:
                samples = list(fp)
        elif args.dataset_name == "strategyqa":
            with open(args.data_filename) as fp:
                samples = json.load(fp)

    model = Generate_Negation(args)
    ExptConfig = GlobalConfig(
        seed = args.seed,
        dataset_name = args.dataset_name
    )

    print(f"The cofig for this experiment = {ExptConfig}")

    datas = []

    for sample_idx, sample in tqdm(enumerate(samples), total = len(samples)):
        if args.dataset_name == "com2sense":
            if args.mode == "normal":
                Q_tilde = model.prompt_tilde(sample["sent"])
        elif args.dataset_name == "csqa2":
            if args.mode == "normal":
                Q_tilde = model.prompt_tilde(sample["question"])
        elif args.dataset_name == "creak":
            if args.mode == "normal":
                Q_tilde = model.prompt_tilde(sample["sentence"])
        elif args.dataset_name == "strategyqa":
            if args.mode == "normal":
                # Q_tilde = model.prompt_tilde(sample["question"])
                Q_general = model.prompt_sent(sample["question"])
                Q_tilde = model.prompt_tilde(Q_general)
                data = {}
                data["Q"] = Q_general
                data["Q_tilde"] = Q_tilde
                data["Q_orig"] = sample["question"]
                data["A"] = sample["answer"]
                datas.append(data)
                # print(f"Q = {sample['question']}")
                # print(f"Q_tilde = {Q_tilde}")
                # print(f"Q general sentence = {Q_general}")


    with open(args.out_filename, "w") as fp:
        json.dump(datas, fp)