llm_engine_example.py 2.18 KB
Newer Older
1
# SPDX-License-Identifier: Apache-2.0
2
3
4
5
"""
This file demonstrates using the `LLMEngine`
for processing prompts with various sampling parameters.
"""
6
7
import argparse

8
from vllm import EngineArgs, LLMEngine, RequestOutput, SamplingParams
9
from vllm.utils import FlexibleArgumentParser
10
11


12
def create_test_prompts() -> list[tuple[str, SamplingParams]]:
13
14
    """Create a list of test prompts with their sampling parameters."""
    return [
Zhuohan Li's avatar
Zhuohan Li committed
15
        ("A robot may not injure a human being",
16
         SamplingParams(temperature=0.0, logprobs=1, prompt_logprobs=1)),
17
18
19
        ("To be or not to be,",
         SamplingParams(temperature=0.8, top_k=5, presence_penalty=0.2)),
        ("What is the meaning of life?",
20
21
22
23
         SamplingParams(n=2,
                        temperature=0.8,
                        top_p=0.95,
                        frequency_penalty=0.1)),
24
25
    ]

26
27

def process_requests(engine: LLMEngine,
28
                     test_prompts: list[tuple[str, SamplingParams]]):
29
    """Continuously process a list of prompts and handle the outputs."""
30
    request_id = 0
31

32
    print('-' * 50)
33
    while test_prompts or engine.has_unfinished_requests():
34
35
        if test_prompts:
            prompt, sampling_params = test_prompts.pop(0)
Zhuohan Li's avatar
Zhuohan Li committed
36
            engine.add_request(str(request_id), prompt, sampling_params)
Zhuohan Li's avatar
Zhuohan Li committed
37
            request_id += 1
38

39
        request_outputs: list[RequestOutput] = engine.step()
40

41
        for request_output in request_outputs:
42
            if request_output.finished:
43
                print(request_output)
44
                print('-' * 50)
45

46
47
48
49
50
51
52
53
54
55
56
57

def initialize_engine(args: argparse.Namespace) -> LLMEngine:
    """Initialize the LLMEngine from the command line arguments."""
    engine_args = EngineArgs.from_cli_args(args)
    return LLMEngine.from_engine_args(engine_args)


def main(args: argparse.Namespace):
    """Main function that sets up and runs the prompt processing."""
    engine = initialize_engine(args)
    test_prompts = create_test_prompts()
    process_requests(engine, test_prompts)
58
59
60


if __name__ == '__main__':
61
    parser = FlexibleArgumentParser(
Zhuohan Li's avatar
Zhuohan Li committed
62
63
        description='Demo on using the LLMEngine class directly')
    parser = EngineArgs.add_cli_args(parser)
64
65
    args = parser.parse_args()
    main(args)