File size: 9,156 Bytes
6fcd376
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
243
244
245
246
247
import argparse
import os
import torch
from pathlib import Path
from tqdm import tqdm
from transformers import AutoTokenizer, AutoModelForCausalLM
import logging
from evalplus.data import (get_human_eval_plus, 
    write_jsonl,
    get_human_eval_plus_hash,
    get_mbpp_plus,
    get_mbpp_plus_hash,
)
from utils import sanitize_solution,check_correctness,get_groundtruth,SUCCESS
from evalplus.eval._special_oracle import MBPP_OUTPUT_NOT_NONE_TASKS
from copy import deepcopy

MAX_TRY = 2

def build_humaneval_instruction(languge: str, question: str):
    return '''You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.

@@ Instruction
Here is the given code to do completion:
```{}
{}
```
Please continue to complete the function with {} programming language. You are not allowed to modify the given code and do the completion only. 

Please return all completed codes in one code block. 
This code block should be in the following format:
```{}
# Your codes here
```

@@ Response
'''.strip().format(languge.lower(), question.strip(),languge.lower(),languge.lower())

build_mbpp_instruction='''You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.

@@ Instruction
Here is the given problem and test examples:
{}

Please use the {} programming language to solve this problem.
Please make sure that your code includes the functions from the test samples and that the input and output formats of these functions match the test samples.

Please return all completed codes in one code block. 
This code block should be in the following format:
```{}
# Your codes here
```

@@ Response
'''

def generate_multi_round(problem, expected_output, example, lang, tokenizer, model, name, flags):
    if flags.dataset=="humaneval":
        prompt = build_humaneval_instruction(lang, example['prompt'])
    elif flags.dataset=="mbpp":
        prompt = build_mbpp_instruction.strip().format(example['prompt'],"python","python")
    inputs = tokenizer.apply_chat_template(
        [{'role': 'user', 'content': prompt }],
        return_tensors="pt"
    ).to(model.device)
    stop_id = tokenizer.convert_tokens_to_ids("<|EOT|>")
    assert isinstance(stop_id, int), "Invalid tokenizer, EOT id not found"
    max_new_tokens=1024
    outputs = model.generate(
        inputs, 
        max_new_tokens=max_new_tokens,
        do_sample=False,
        pad_token_id=tokenizer.eos_token_id,
        eos_token_id=tokenizer.eos_token_id,
        temperature=0,
    )
    output = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)
    solution = {k:v for k,v in example.items()}
    solution["solution"]=output
    sanitized_solution = sanitize_solution(deepcopy(solution),flags.eofs)
    attempt = 1
    judge = False
    modify = False
    code = sanitized_solution["solution"]
    while attempt==1 or sanitized_solution["solution"]!="":
        args = (
            flags.dataset,
            0,
            problem,
            sanitized_solution["solution"],
            expected_output,
            flags.version,
            True,  # fast_check
            example["task_id"]+f'_{attempt}',
            flags.min_time_limit,
            flags.gt_time_limit_factor,
        )
        result = check_correctness(*args)
        if flags.version=="base" and result["base"][0]==SUCCESS:
            code = sanitized_solution["solution"]
            if attempt==2:
                modify = True
            judge = True
            break
        elif flags.version=="plus" and result["plus"][0]==result["base"][0]==SUCCESS:
            code = sanitized_solution["solution"]
            if attempt==2:
                modify = True
            judge = True
            break
        else:
            attempt += 1    
            if attempt > MAX_TRY:
                code = sanitized_solution["solution"]
                break
            execution_feedback=""
            if flags.version=="base":
                execution_feedback=result["base"][2]
            elif flags.version=="plus":
                if result["base"][0]!=SUCCESS:
                    execution_feedback+=result["base"][2]
                    if "The results aren't as expected." in execution_feedback:
                        if result["plus"][0]!=SUCCESS:
                            execution_feedback+="\n"+result["plus"][2]
                else:
                    execution_feedback=result["plus"][2]
            prompt +="""
{}

@@ Instruction
Execution result: 
{}
""".format(solution["solution"],execution_feedback)
            inputs = tokenizer.apply_chat_template(
                [{'role': 'user', 'content': prompt }],
                return_tensors="pt"
            ).to(model.device)

            outputs = model.generate(
                inputs, 
                max_new_tokens=max_new_tokens,
                do_sample=False,
                pad_token_id=tokenizer.eos_token_id,
                eos_token_id=tokenizer.eos_token_id,
                temperature=0,
            )
            output = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)
            solution = {k:v for k,v in example.items()}
            solution["solution"]=output 
            sanitized_solution = sanitize_solution(deepcopy(solution),flags.eofs)

    return code,judge,modify


def gen_solution(args):
    os.environ["TOKENIZERS_PARALLELISM"] = "false"
    fail_list=[]
    model_path = args.model
    logging.info(f"model:{model_path}")
    model_name =model_path.replace("/", "_")
    lang = "python"
    os.makedirs(os.path.join(args.output_path,model_name),exist_ok=True)
    output_file = os.path.join(args.output_path,model_name,f"multiround_{args.dataset}_{args.version}_solutions-sanitized.jsonl")
    if os.path.exists(output_file):
        logging.info(f"Old sample jsonl file exists, remove it. {output_file}")
        os.remove(output_file)

    tokenizer = AutoTokenizer.from_pretrained(model_path)
    logging.info("load tokenizer {} from {} over.".format(tokenizer.__class__, model_path))
    model = AutoModelForCausalLM.from_pretrained(
        model_path,
        torch_dtype=torch.bfloat16,
        device_map="auto",
    )
    model.eval()

    modelname=model_path.replace("/", "_")
    if args.dataset=="humaneval":
        problems = get_human_eval_plus()
        examples = problems.items()
        dataset_hash = get_human_eval_plus_hash()
        expected_outputs = get_groundtruth(problems, dataset_hash, [])
    else:
        problems = get_mbpp_plus()
        examples = problems.items()
        dataset_hash = get_mbpp_plus_hash()
        expected_outputs = get_groundtruth(
                problems,
                dataset_hash,
                MBPP_OUTPUT_NOT_NONE_TASKS,
            )
    logging.info("Read {} examples for evaluation over.".format(len(examples)))
    a,b = 0,0
    total_modify = 0
    for task_id,example in tqdm(examples, desc='Generating'):
        problem = problems[task_id]
        expected_output = expected_outputs[task_id]
        code,judge,modify = generate_multi_round(problem,expected_output,example, lang, tokenizer, model, modelname,args)
        gen_sample=[dict(task_id=task_id, solution=code)]
        write_jsonl(output_file, gen_sample ,append=True)

        if modify:
            total_modify += 1
        if judge:
            a += 1
        else:
            b += 1
            fail_list.append(task_id)
        result = a/(a+b)
        print ("pass num :",a)
        print ("total num:",a+b)
        print ('pass rate: '+str(result))
        print ("num modify: "+str(total_modify))
        print ("judge:",judge)
        print ('modify: '+str(modify))
        print ("fail list:",fail_list)


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--model', type=str, help="model path")
    parser.add_argument('--output_path', type=str, help="output path", default="./multiround_output")
    parser.add_argument('--log_file', type=str, help="log file name", default="gen_humaneval_plus_solution_singleround.log")
    parser.add_argument("--min-time-limit", default=1, type=float)
    parser.add_argument("--gt-time-limit-factor", default=4.0, type=float)
    parser.add_argument(
        "--version", required=True, type=str, choices=["base", "plus"]
    )
    parser.add_argument(
        "--dataset", required=True, type=str, choices=["humaneval", "mbpp"]
    )

    args = parser.parse_args()
    args.eofs=None

    model_name = args.model.replace("/", "_")
    os.makedirs(os.path.join(args.output_path,model_name),exist_ok=True)
    logfile=os.path.join(args.output_path,model_name,args.log_file)
    logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - \n%(message)s')
    file_handler = logging.FileHandler(logfile)
    file_handler.setLevel(logging.INFO)
    formatter = logging.Formatter('%(asctime)s - %(levelname)s - \n%(message)s')
    file_handler.setFormatter(formatter)  
    logging.getLogger().addHandler(file_handler)

    gen_solution(args)