A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 17, 01:45 UTC
[Misc] Keep only one implementation of the create_dummy_prompt function. (#4716)
Allen.Dou committed
May 10, 2024 at 12:42 UTC
e965d4618430db94fed73d4305c44c9147e47150
1 file changed
+4
-32
tests/test_sequence.py
+4
-32
@@ -1,36 +1,8 @@
1
-import time
2
-from typing import Optional
3
-
1
import pytest
2
6
-from vllm import SamplingParams
7
-from vllm.lora.request import LoRARequest
8
-from vllm.sequence import (SamplerOutput, Sequence, SequenceData,
9
- SequenceGroup, SequenceGroupOutput, SequenceOutput)
10
-
11
-
12
-def create_dummy_prompt(
13
- request_id: str,
14
- prompt_length: int,
15
- block_size: Optional[int] = None,
16
- lora_request: Optional[LoRARequest] = None,
17
- use_beam_search: bool = False,
18
- best_of: int = 1,
19
-) -> SequenceGroup:
20
- if not block_size:
21
- block_size = prompt_length
22
-
23
- # Create dummy prompt sequence with tokens 0...block_size-1
24
- # and prompt "0 ... block_size".
25
- prompt_tokens = list(range(prompt_length))
26
- prompt_str = " ".join([str(t) for t in prompt_tokens])
27
- prompt = Sequence(int(request_id), prompt_str, prompt_tokens, block_size)
28
- seq_group = SequenceGroup(
29
- request_id, [prompt],
30
- SamplingParams(use_beam_search=use_beam_search, best_of=best_of),
31
- time.time(), lora_request)
32
-
33
- return seq_group
3
+from tests.core.utils import create_dummy_prompt
4
+from vllm.sequence import (SamplerOutput, SequenceData, SequenceGroupOutput,
5
+ SequenceOutput)
6
7
8
@pytest.fixture
@@ -102,7 +74,7 @@ def test_sequence_data_prefill():
74
75
76
def test_sequence_group_stage():
105
- seq_group = create_dummy_prompt("1", 12)
77
+ _, seq_group = create_dummy_prompt("1", 12)
78
assert seq_group.is_prefill() is True
79
seq_group.update_num_computed_tokens(6)
80
assert seq_group.is_prefill() is True