File size: 5,193 Bytes
2a26d3b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import argparse
import json
import os
import re
from pathlib import Path

import torch
from tqdm import tqdm

data_abs_dir = Path(__file__).parent / "data"

from human_eval.evaluation import evaluate_functional_correctness
from transformers import AutoModelForCausalLM, AutoTokenizer


def read_test_examples(data_path: str):
    def format_test_example(q, tests, code: str = None):
        prompt = ">>> Problem:\n{}\n>>> Test Cases:\n{}\n".format(
            q.strip(), "\n".join(tests)
        )
        if code:
            code = code.replace("\r", "").replace("\t", "    ")
            prompt += "\n>>> Code:\n```python\n{}\n```".format(code)
        return prompt

    examples = [json.loads(x) for x in open(data_path)]
    print("Read all {} examples from {} over!".format(len(examples), data_path))

    # test_cases
    examples_str = []
    for i in range(1, 4):
        ex = examples[i]
        q, test, code = ex["text"], ex["test_list"], ex["code"]
        ex_prompt = format_test_example(q, test, code)
        example_prompt = "- Example {}:\n{}".format(i, ex_prompt)
        examples_str += [example_prompt]

    for i in range(10, 510):
        ex = examples[i]
        q, test, code = ex["text"], ex["test_list"], ex["code"]

        prompt = format_test_example(q, test, code=None)

        prompt_with_shots = """
Please refer the given examples and generate a python function for my problem.
Examples are listed as follows:
{}

Here is my problem:
{}
""".strip().format(
            "\n\n".join(examples_str), prompt
        )
        yield {"task_id": ex["task_id"], "prompt": prompt_with_shots}


def convert_for_evaluation(example):
    gpt_completion = example["gpt_completion"]
    generation = gpt_completion
    try:
        code_block: str = re.findall(
            f"```python\n(.*?)```", gpt_completion, re.DOTALL | re.IGNORECASE
        )[0]
        generation = code_block
    except Exception as ex:
        print("Failed to extract codeblock:\n{}".format(gpt_completion))

    example["generation"] = generation
    return example


def generate_one(example, tokenizer, model):
    prompt = example["prompt"]
    inputs = tokenizer.apply_chat_template(
        [{"role": "user", "content": prompt}],
        return_tensors="pt",
        add_generation_prompt=True,
    ).to(model.device)
    #
    # stop_id = tokenizer.convert_tokens_to_ids("<|EOT|>")
    # assert isinstance(stop_id, int), "Invalid tokenizer, EOT id not found"
    stop_id = tokenizer.eos_token_id
    outputs = model.generate(
        inputs,
        max_new_tokens=512,
        do_sample=False,
        # top_p=0.95,
        # temperature=temperature,
        pad_token_id=stop_id,
        eos_token_id=stop_id,
    )

    output = tokenizer.decode(outputs[0][len(inputs[0]) :], skip_special_tokens=True)
    # print(output)
    example["gpt_completion"] = output
    return convert_for_evaluation(example)


def generate_main(args):
    model_name_or_path = args.model
    saved_path = args.output_path
    temp_dir = args.temp_dir
    os.makedirs(temp_dir, exist_ok=True)
    problem_file = os.path.join(data_abs_dir, f"mbpp.jsonl")

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

    examples = list(read_test_examples(problem_file))
    print("Read {} examples for evaluation over.".format(len(examples)))

    generated_examples = []
    for ex in tqdm(examples, desc="Generating"):
        gen_example = generate_one(ex, tokenizer, model)
        generated_examples.append(gen_example)
        print("Generate {}/{} over...".format(len(generated_examples), len(examples)))

    print("Generate all over!!!")
    with open(saved_path, "w", encoding="utf-8") as fw:
        for ex in generated_examples:
            fw.write(json.dumps(ex) + "\n")
        print(
            "Save {} processed examples into {} over!".format(
                len(generated_examples), saved_path
            )
        )

    result = evaluate_functional_correctness(
        input_file=saved_path,
        tmp_dir=temp_dir,
        problem_file=os.path.join(data_abs_dir, f"mbpp_test.jsonl"),
        language="python",
        is_mbpp=True,
    )
    print(result, model_name_or_path)
    pass


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--model",
        type=str,
        help="model name or path",
        default="/data0/pretrained-models/Qwen2-7B-Instruct",
    )
    parser.add_argument(
        "--output_path",
        type=str,
        help="output path of your generation",
        default="/home/qyhuang/DeepSeek-Coder/outputs/qwen2-mbpp.json",
    )
    parser.add_argument(
        "--temp_dir", type=str, help="temp dir for evaluation", default="tmp"
    )
    args = parser.parse_args()

    os.environ["TOKENIZERS_PARALLELISM"] = "false"
    generate_main(args)
    pass