JavaBench/inference.py at main · java-bench/JavaBench · 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
import random
import argparse
import os
import itertools
import logging
from tqdm import tqdm
from fastchat.model import load_model, get_conversation_template, add_model_args
from app.prompt.template import complete_template
from app.static_analyzer.class_compose_tool import get_todo_methods, replace_method, retain_todo_method
from app.util.io import extract_code, stream_jsonl, write_jsonl
from langchain_openai.chat_models import ChatOpenAI

def inference(args):
    is_openai = args.model_path.startswith("gpt")
    if is_openai:
        model = ChatOpenAI(model=args.model_path, temperature=args.temperature)
    else:
        model, tokenizer = load_model(
            args.model_path,
            device=args.device,
            num_gpus=args.num_gpus,
            max_gpu_memory=args.max_gpu_memory,
            load_8bit=args.load_8bit,
            cpu_offloading=args.cpu_offloading,
            revision=args.revision,
            debug=args.debug,
        )
        tokenizer.pad_token = tokenizer.eos_token

    def query(code, code_context):
        lc_messages = complete_template.format_messages(
            code_context=code_context,
            code=code,
        )

        if is_openai:
            prompt = lc_messages[0].content + "\n" + lc_messages[1].content
            outputs = model.invoke(lc_messages).content
        else:
            conv = get_conversation_template(args.model_path)
            if "{system_message}" in conv.system_template:
                conv.system_message = lc_messages[0].content
            else:
                conv.append_message(conv.roles[0], lc_messages[0].content)
            conv.append_message(conv.roles[0], lc_messages[1].content)
            conv.append_message(conv.roles[1], None)
            prompt = conv.get_prompt()

            # Run inference
            inputs = tokenizer([prompt], return_tensors="pt").to(args.device)
            output_ids = model.generate(
                **inputs,
                do_sample=True if args.temperature > 1e-5 else False,
                temperature=args.temperature,
                repetition_penalty=args.repetition_penalty,
                max_new_tokens=args.max_new_tokens,
            )
            if model.config.is_encoder_decoder:
                output_ids = output_ids[0]
            else:
                output_ids = output_ids[0][len(inputs["input_ids"][0]) :]
            outputs = tokenizer.decode(
                output_ids, skip_special_tokens=True, spaces_between_special_tokens=False
            )
        return prompt, outputs

    tasks = list(stream_jsonl(args.data))
    samples = list(stream_jsonl(args.output)) if os.path.exists(args.output) else []
    for task, _ in tqdm(itertools.islice(itertools.product(tasks, range(args.num_sample)), len(samples), None), total=len(tasks) * args.num_sample, initial=len(samples)):
        if args.mode == "holistic":
            prompt, outputs = query(task["code"], task["code_context"])
            samples.append(dict(
                task_id=task["task_id"],
                target=task["target"],
                prompt=prompt,
                completion=outputs,
            ))
        elif args.mode == "independent":
            result = task["code"]
            mediate = []

            todo_methods = get_todo_methods(result)
            progress = tqdm(todo_methods)
            for todo_method in progress:
                progress.set_description(f"{todo_method['name']} {todo_method['seq']}")
                source = retain_todo_method(task["code"], todo_method["name"], todo_method["seq"])
                prompt, outputs = query(source, task["code_context"])
                result = replace_method(result, extract_code(outputs), todo_method["name"], todo_method["seq"])
                new_mediate = dict(
                    name=todo_method["name"],
                    seq=todo_method["seq"],
                    prompt=prompt,
                    completion=outputs,
                )
                mediate.append(new_mediate)
                logging.info(f"{todo_method['name']} {todo_method['seq']} {new_mediate}")

            samples.append(dict(
                task_id=task["task_id"],
                target=task["target"],
                completion=result,
                mediate=mediate,
            ))
        elif args.mode == "incremental":
            result = task["code"]
            mediate = []

            todo_methods = get_todo_methods(result)
            if args.incremental_mode == "rev":
                todo_methods = reversed(todo_methods)
            elif args.incremental_mode == "rand":
                random.shuffle(todo_methods)
            progress = tqdm(todo_methods)
            for todo_method in progress:
                progress.set_description(f"{todo_method['name']} {todo_method['seq']}")
                source = retain_todo_method(result, todo_method["name"], todo_method["seq"])
                prompt, outputs = query(source, task["code_context"])
                result = replace_method(result, extract_code(outputs), todo_method["name"], todo_method["seq"])
                new_mediate = dict(
                    name=todo_method["name"],
                    seq=todo_method["seq"],
                    prompt=prompt,
                    completion=outputs,
                )
                mediate.append(new_mediate)
                logging.info(f"{todo_method['name']} {todo_method['seq']} {new_mediate}")

            samples.append(dict(
                task_id=task["task_id"],
                target=task["target"],
                completion=result,
                mediate=mediate,
            ))
        if os.path.dirname(args.output):
            os.makedirs(os.path.dirname(args.output), exist_ok=True)
        write_jsonl(args.output, samples)


if __name__ == "__main__":
    os.makedirs("logs", exist_ok=True)
    logging.basicConfig(format='%(levelname)s %(asctime)s %(message)s', datefmt='%m/%d/%Y %I:%M:%S', filename=f"logs/inference-{os.getpid()}.log", filemode="w", level=logging.DEBUG)

    parser = argparse.ArgumentParser()
    add_model_args(parser)
    parser.add_argument(
        "--mode",
        type=str,
        choices=["holistic", "independent", "incremental"],
        default="holistic",
    )

    parser.add_argument("--data", type=str, required=True)
    parser.add_argument("--num-sample", type=int, default=10)
    parser.add_argument("--output", type=str, required=True)
    parser.add_argument("--incremental-mode", type=str, choices=["seq", "rev", "rand"], default="seq")

    parser.add_argument("--temperature", type=float, default=0.2)
    parser.add_argument("--repetition_penalty", type=float, default=1.0)
    parser.add_argument("--max-new-tokens", type=int, default=4096)
    parser.add_argument("--debug", action="store_true")
    args = parser.parse_args()

    inference(args)