PUB/eval_openAI_ppa.py at main · meetdoshi90/PUB · 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
import pandas as pd
import os
from datetime import datetime
from pathlib import Path
from typing import List, Optional, Union
import csv
import openai
from tqdm.auto import tqdm
from pragmatics.eval_tasks.prompt_script import get_data
from datasets import Dataset
import ast
from sklearn.metrics import precision_score, recall_score, f1_score, accuracy_score
import time
import json

openai.api_key = ""


def evaluate(dataset,metadata):
    k = 0
    data_len = len(dataset)
    outputs = []
    file_name = "output_zero.json"
    print(data_len)
    i=0
    num_options = len(dataset['answer_options'][0])
    with tqdm(total=data_len) as pbar:
        while i<data_len:

            try:
                response = openai.Completion.create(
                    model="gpt-3.5-turbo-instruct",
                    #messages=[
                    #    {"role": "system", "content": "You are a helpful assistant."},
                    #    {"role": "user", "content": dataset['wrapped'][i]}
                    #],
                    prompt = dataset['wrapped'][i],
                    max_tokens=1,
                    temperature=1.0,
                    logprobs=5,
                )
            except Exception as e:
                if e:
                    print(e)
                    print('Timeout error, retrying...')
                    time.sleep(180)
                else:
                    raise e
            if response and response["choices"] and response["choices"][0] and response["choices"][0]["logprobs"] and response["choices"][0]["logprobs"]["top_logprobs"] :
                result_dict = {}
                logprobs = dict(response["choices"][0]["logprobs"]["top_logprobs"][0])
                options = ast.literal_eval(dataset['options'][i])
                rrd = {'prompt':dataset['wrapped'][i],'probs':logprobs,'correct_answer':chr(options.index(dataset['correct answer'][i])+ord('A'))}
                with open(file_name, 'a') as file:
                    json.dump(rrd, file)
                    file.write('\n')
                for key,value in logprobs.items():
                    cleaned_key = key.strip().upper()
                    if cleaned_key.isalpha():
                        if cleaned_key in result_dict:
                            result_dict[cleaned_key] = max([result_dict[cleaned_key],value])
                        else:
                            result_dict[cleaned_key] = value

                outputs.append(max(result_dict, key=result_dict.get))
                i+=1
                pbar.update(1)
            else:
                k+=1
                generated_response = "ERROR"
                print(generated_response)

    print(k)
    df = dataset.to_pandas()
    options = [ast.literal_eval(o) for o in df['options']]
    correct_options = [chr(options[i].index(df['correct answer'][i])+ord('A')) for i in range(len(df['correct answer']))]
    df['correct_options'] = correct_options
    predictions = [outputs[i][0] for i in range(len(outputs))]
    df['generate_predictions'] = predictions
    accuracy = accuracy_score(df['correct_options'], df['generate_predictions'])
    precision = precision_score(df['correct_options'], df['generate_predictions'],average='macro')
    recall = recall_score(df['correct_options'], df['generate_predictions'],average='macro')
    f1_macro = f1_score(df['correct_options'], df['generate_predictions'],average='macro')
    results_dict = {
    'precision':precision,
    'recall':recall,
    'f1_macro':f1_macro,
    'accuracy':accuracy
    }

    if not os.path.exists("gpt3_results_zero_shot"):
        os.makedirs("gpt3_results_zero_shot")
    absolute_metric_path = os.path.join("gpt3_results_zero_shot",str(metadata['task']) + ".csv")
    absolute_result_path = os.path.join("gpt3_results_zero_shot","data_" +str(metadata['task']) + ".csv")
    df.to_csv(absolute_result_path, index=True)
    with open(absolute_metric_path, 'w') as json_file:
        json.dump(results_dict, json_file)

if __name__=="__main__":
    tasks = [0,1,2,3,4,5,6,7,8,9,10,13,14,3,11,15]
    # tasks = [4]
    metadata= {
        'task':0,
        'model':'gpt2',
        'input_dir':'./pragmatics/global_datasets/',
        'model_path':'/raid/nlp/pranavg/iclr/pragmatics/models/llama-2-13b-chat-hf/',
        'prompt':'zero_shot', # zero_shot to few_shot
        'run':0,
        'output_dir':'/raid/nlp/pranavg/iclr/Results',
        'eval_method':'mcqa', #mcqa or harness or ppa
        'k':0 # 0,3,5
    }
    for t in tasks:
        print(metadata['model'])
        print(metadata['prompt'])
        print(metadata['k'])
        print(metadata['eval_method'])
        print(t)

        metadata['task'] = t
        data_file_name = f"task_{metadata['task']}.csv"
        data_path = os.path.join(metadata['input_dir'],data_file_name)
        gdata = pd.read_csv(data_path)
        dataset = get_data(gdata,metadata)
        dataset_pd = dataset
        # dataset = dataset[:5]
        dataset = Dataset.from_pandas(dataset)
        print("-------")
        print(type(dataset['wrapped']),len(dataset))
        print("-----")
        evaluate(dataset,metadata)