benchmark_prefix_caching.py 10.4 KB
Newer Older
1
# SPDX-License-Identifier: Apache-2.0
2
3
4
5
6
7
8
9
10
11
12
13
"""
Benchmark the efficiency of prefix caching.

This script allows you to benchmark the performance of
a model with and without prefix caching using either fixed prompts
or prompts sampled from the ShareGPT dataset.

Fixed example usage:
    python benchmark_prefix_caching.py \
        --model meta-llama/Llama-2-7b-chat-hf \
        --enable-prefix-caching \
        --num-prompts 1 \
14
15
        --repeat-count 100 \
        --input-length-range 128:256
16
17
18
19
20
21
22
23
24
25
26
27
28
29

ShareGPT example usage:
    # This command samples 20 prompts with input lengths
    # between 128 and 256 tokens from the ShareGPT dataset,
    # then replicates each prompt 5 times.
    python benchmark_prefix_caching.py \
        --model meta-llama/Llama-2-7b-chat-hf \
        --dataset-path /path/to/ShareGPT_V3_unfiltered_cleaned_split.json \
        --enable-prefix-caching \
        --num-prompts 20 \
        --repeat-count 5 \
        --input-length-range 128:256
"""

30
import dataclasses
31
32
import json
import random
33
import time
34
from typing import Optional
35
36

from transformers import PreTrainedTokenizerBase
37

38
from vllm import LLM, SamplingParams
39
from vllm.engine.arg_utils import EngineArgs
40
from vllm.utils import FlexibleArgumentParser
41

42
43
44
45
46
try:
    from vllm.transformers_utils.tokenizer import get_tokenizer
except ImportError:
    from backend_request_func import get_tokenizer

47
PROMPT = "You are a helpful assistant in recognizes the content of tables in markdown format. Here is a table as fellows. You need to answer my question about the table.\n# Table\n|Opening|Opening|Sl. No.|Film|Cast|Director|Music Director|Notes|\n|----|----|----|----|----|----|----|----|\n|J A N|9|1|Agni Pushpam|Jayabharathi, Kamalahasan|Jeassy|M. K. Arjunan||\n|J A N|16|2|Priyamvada|Mohan Sharma, Lakshmi, KPAC Lalitha|K. S. Sethumadhavan|V. Dakshinamoorthy||\n|J A N|23|3|Yakshagaanam|Madhu, Sheela|Sheela|M. S. Viswanathan||\n|J A N|30|4|Paalkkadal|Sheela, Sharada|T. K. Prasad|A. T. Ummer||\n|F E B|5|5|Amma|Madhu, Srividya|M. Krishnan Nair|M. K. Arjunan||\n|F E B|13|6|Appooppan|Thikkurissi Sukumaran Nair, Kamal Haasan|P. Bhaskaran|M. S. Baburaj||\n|F E B|20|7|Srishti|Chowalloor Krishnankutty, Ravi Alummoodu|K. T. Muhammad|M. S. Baburaj||\n|F E B|20|8|Vanadevatha|Prem Nazir, Madhubala|Yusufali Kechery|G. Devarajan||\n|F E B|27|9|Samasya|Madhu, Kamalahaasan|K. Thankappan|Shyam||\n|F E B|27|10|Yudhabhoomi|K. P. Ummer, Vidhubala|Crossbelt Mani|R. K. Shekhar||\n|M A R|5|11|Seemantha Puthran|Prem Nazir, Jayabharathi|A. B. Raj|M. K. Arjunan||\n|M A R|12|12|Swapnadanam|Rani Chandra, Dr. Mohandas|K. G. George|Bhaskar Chandavarkar||\n|M A R|19|13|Thulavarsham|Prem Nazir, sreedevi, Sudheer|N. Sankaran Nair|V. Dakshinamoorthy||\n|M A R|20|14|Aruthu|Kaviyoor Ponnamma, Kamalahasan|Ravi|G. Devarajan||\n|M A R|26|15|Swimming Pool|Kamal Haasan, M. G. Soman|J. Sasikumar|M. K. Arjunan||\n\n# Question\nWhat' s the content in the (1,1) cells\n"  # noqa: E501
48
49


50
def test_prefix(llm=None, sampling_params=None, prompts=None):
51
    start_time = time.time()
52
53

    llm.generate(prompts, sampling_params=sampling_params)
54
55
56
57
58

    end_time = time.time()
    print(f"cost time {end_time - start_time}")


59
60
61
62
63
64
65
@dataclasses.dataclass
class Request:
    prompt: str
    prompt_len: int
    output_len: int


66
67
def sample_tokens(tokenizer: PreTrainedTokenizerBase,
                  length: int) -> list[int]:
68
    vocab = tokenizer.get_vocab()
69
70
    all_special_ids = set(tokenizer.all_special_ids)

71
    # Remove the special tokens.
72
73
74
75
    return random.choices(
        [v for k, v in vocab.items() if k not in all_special_ids],
        k=length,
    )
76
77
78


def sample_requests_from_dataset(
79
80
81
    dataset_path: str,
    num_requests: int,
    tokenizer: PreTrainedTokenizerBase,
82
    input_length_range: tuple[int, int],
83
    fixed_output_len: Optional[int],
84
) -> list[Request]:
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
    if fixed_output_len is not None and fixed_output_len < 4:
        raise ValueError("output_len too small")

    # Load the dataset.
    with open(dataset_path) as f:
        dataset = json.load(f)
    # Filter out the conversations with less than 2 turns.
    dataset = [data for data in dataset if len(data["conversations"]) >= 2]
    # Only keep the first two turns of each conversation.
    dataset = [(data["conversations"][0]["value"],
                data["conversations"][1]["value"]) for data in dataset]

    # Shuffle the dataset.
    random.shuffle(dataset)

    min_len, max_len = input_length_range
101
    assert min_len >= 0 and max_len >= min_len, "input_length_range too small"
102
103

    # Filter out sequences that are too long or too short
104
    filtered_requests: list[Request] = []
105

106
    for i in range(len(dataset)):
107
        if len(filtered_requests) == num_requests:
108
109
110
            break

        # Tokenize the prompts and completions.
111
112
        prompt_token_ids = tokenizer(dataset[i][0]).input_ids
        prompt = tokenizer.decode(prompt_token_ids)
113
114
115
        completion = dataset[i][1]
        completion_token_ids = tokenizer(completion).input_ids
        prompt_len = len(prompt_token_ids)
116
117
        output_len = (len(completion_token_ids)
                      if fixed_output_len is None else fixed_output_len)
118
        if min_len <= prompt_len <= max_len:
119
120
121
122
123
124
125
126
            filtered_requests.append(Request(prompt, prompt_len, output_len))

    return filtered_requests


def sample_requests_from_random(
    num_requests: int,
    tokenizer: PreTrainedTokenizerBase,
127
    input_length_range: tuple[int, int],
128
129
    fixed_output_len: Optional[int],
    prefix_len: int,
130
) -> list[Request]:
131

132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
    requests = []
    prefix_token_ids = sample_tokens(tokenizer, prefix_len)
    min_len, max_len = input_length_range

    for i in range(num_requests):
        unique_part_token_ids = sample_tokens(
            tokenizer,
            random.randint(min_len - prefix_len, max_len - prefix_len))
        prompt_token_ids = prefix_token_ids + unique_part_token_ids
        prompt = tokenizer.decode(prompt_token_ids)
        prompt_len = len(prompt_token_ids)
        assert (min_len <= prompt_len <= max_len
                ), f"prompt_len {prompt_len} out of range {min_len}:{max_len}"
        requests.append(Request(prompt, prompt_len, fixed_output_len))
    return requests
147
148


149
def repeat_and_sort_requests(requests: list[Request],
150
                             repeat_count: int,
151
                             sort: bool = False) -> list[str]:
152
153
154
155
156
    repeated_requests = requests * repeat_count
    if sort:
        repeated_requests.sort(key=lambda x: x[1])
    else:
        random.shuffle(repeated_requests)
157
    return [req.prompt for req in repeated_requests]
158
159


160
def main(args):
161
162
    tokenizer = get_tokenizer(args.model, trust_remote_code=True)
    input_length_range = tuple(map(int, args.input_length_range.split(':')))
163
    random.seed(args.seed)
164
    if args.dataset_path is not None:
165
166
167
168
        if args.prefix_len > 0:
            raise ValueError("prefix-len is not supported when "
                             "dataset-path is provided.")
        print(f"Start to sample {args.num_prompts} prompts "
Cody Yu's avatar
Cody Yu committed
169
              f"from {args.dataset_path}")
170
        filtered_requests = sample_requests_from_dataset(
171
172
173
174
175
176
177
            dataset_path=args.dataset_path,
            num_requests=args.num_prompts,
            tokenizer=tokenizer,
            input_length_range=input_length_range,
            fixed_output_len=args.output_len,
        )
    else:
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
        print(f"Start to sample {args.num_prompts} prompts from random")
        filtered_requests = sample_requests_from_random(
            num_requests=args.num_prompts,
            tokenizer=tokenizer,
            input_length_range=input_length_range,
            fixed_output_len=args.output_len,
            prefix_len=args.prefix_len,
        )

    # Print some helpful stats of the requests.
    print(f"Sampled {len(filtered_requests)} requests.")
    prompt_lens = [req.prompt_len for req in filtered_requests]
    print(f"Average input length: {sum(prompt_lens) / len(prompt_lens)}")
    print(f"P50 input length: {sorted(prompt_lens)[len(prompt_lens) // 2]}")
    print(f"Min Prompt Length: {min(prompt_lens)}")
    print(f"Max Prompt Length: {max(prompt_lens)}")
194

195
196
197
    engine_args = EngineArgs.from_cli_args(args)

    llm = LLM(**dataclasses.asdict(engine_args))
198

199
200
201
    sampling_params = SamplingParams(temperature=0,
                                     max_tokens=args.output_len,
                                     detokenize=not args.disable_detokenize)
202

203
204
    print("Testing filtered requests")
    prompts = repeat_and_sort_requests(filtered_requests,
205
206
207
                                       repeat_count=args.repeat_count,
                                       sort=args.sort)

208
209
210
211
212
213
214
215
216
    print("------start generating------")
    test_prefix(
        llm=llm,
        prompts=prompts,
        sampling_params=sampling_params,
    )


if __name__ == "__main__":
217
    parser = FlexibleArgumentParser(
218
219
220
221
222
223
        description=
        'Benchmark the performance with or without automatic prefix caching.')
    parser.add_argument("--dataset-path",
                        type=str,
                        default=None,
                        help="Path to the dataset.")
224
    parser.add_argument('--output-len', type=int, default=10)
225
226
    parser.add_argument('--num-prompts',
                        type=int,
227
                        required=True,
228
229
230
                        help="Number of the prompts sampled from dataset")
    parser.add_argument('--repeat-count',
                        type=int,
231
                        default=1,
232
233
234
235
236
237
                        help='Number of times to repeat each prompt')
    parser.add_argument('--sort',
                        action='store_true',
                        help='Sort prompts by input length')
    parser.add_argument('--input-length-range',
                        type=str,
238
                        required=True,
239
240
                        help='Range of input lengths for sampling prompts,'
                        'specified as "min:max" (e.g., "128:256").')
241
242
243
244
245
246
247
248
249
    parser.add_argument(
        "--prefix-len",
        type=int,
        default=0,
        help="Specifies the length of a common prefix to be "
        "added to the input prompt. The input-length-range will "
        "subtract this length when filtering prompts. Only used "
        "when dataset-path is not provided.",
    )
250
251
252
253
254
255
    parser.add_argument(
        '--disable-detokenize',
        action='store_true',
        help=("Do not detokenize responses (i.e. do not include "
              "detokenization time in the latency measurement)"),
    )
256
257

    parser = EngineArgs.add_cli_args(parser)
258
259
    args = parser.parse_args()
    main(args)