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 14, 05:30 UTC
[V0 Deprecation] Remove V0 tests in test_sequence.py (#25088)
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Woosuk Kwon committed
Sep 17, 2025 at 11:08 UTC
9d442b7c48288d6a65cbaca1bba10392523fe94d
1 file changed
+1
-96
tests/test_sequence.py
+1
-96
@@ -1,104 +1,9 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
-import pytest
4
import torch
5
7
-from vllm.model_executor.layers.sampler import SamplerOutput
8
-from vllm.sequence import (CompletionSequenceGroupOutput, IntermediateTensors,
9
- SequenceData, SequenceOutput)
10
-
11
-from .core.utils import create_dummy_prompt
12
-
13
-
14
-@pytest.fixture
15
-def sample_outputs():
16
- return [
17
- CompletionSequenceGroupOutput(samples=[
18
- SequenceOutput(parent_seq_id=0, output_token=i, logprobs={})
19
- ],
20
- prompt_logprobs=None) for i in range(5)
21
- ]
22
-
23
-
24
-@pytest.fixture
25
-def sampler_output(sample_outputs):
26
- return SamplerOutput(outputs=sample_outputs)
27
-
28
-
29
-def test_sampler_output_initialization(sampler_output, sample_outputs):
30
- assert len(sampler_output) == len(sample_outputs)
31
- assert sampler_output.sampled_token_probs is None
32
- assert sampler_output.sampled_token_ids is None
33
-
34
-
35
-def test_sampler_output_getitem(sampler_output, sample_outputs):
36
- assert sampler_output[2] == sample_outputs[2]
37
-
38
-
39
-def test_sampler_output_setitem(sampler_output):
40
- new_output = CompletionSequenceGroupOutput(samples=[
41
- SequenceOutput(parent_seq_id=0, output_token=99, logprobs={})
42
- ],
43
- prompt_logprobs=None)
44
- sampler_output[2] = new_output
45
- assert sampler_output[2] == new_output
46
-
47
-
48
-def test_sampler_output_len(sampler_output, sample_outputs):
49
- assert len(sampler_output) == len(sample_outputs)
50
-
51
-
52
-def test_sampler_output_eq(sample_outputs):
53
- sampler_output1 = SamplerOutput(outputs=sample_outputs)
54
- sampler_output2 = SamplerOutput(outputs=sample_outputs.copy())
55
- sampler_output3 = SamplerOutput(outputs=sample_outputs[:-1])
56
- assert sampler_output1 == sampler_output2
57
- assert sampler_output1 != sampler_output3
58
-
59
-
60
-def test_sequence_data_prefill():
61
- seq_data = SequenceData.from_seqs([1, 2, 3, 4])
62
- assert seq_data.get_num_uncomputed_tokens() == 4
63
- assert seq_data.get_num_computed_tokens() == 0
64
- # advance by 2
65
- seq_data.update_num_computed_tokens(2)
66
- assert seq_data.get_num_uncomputed_tokens() == 2
67
- assert seq_data.get_num_computed_tokens() == 2
68
-
69
- # advance by 1
70
- seq_data.update_num_computed_tokens(1)
71
- assert seq_data.get_num_uncomputed_tokens() == 1
72
- assert seq_data.get_num_computed_tokens() == 3
73
-
74
- # append tokens and reset, simulating recompute
75
- seq_data.append_token_id(1, logprob=0.0)
76
- seq_data.reset_state_for_recompute()
77
- assert seq_data.get_num_uncomputed_tokens() == 5
78
- assert seq_data.get_num_computed_tokens() == 0
79
-
80
-
81
-def test_sequence_group_stage():
82
- _, seq_group = create_dummy_prompt("1", 12)
83
- assert seq_group.is_prefill() is True
84
- seq_group.update_num_computed_tokens(6)
85
- assert seq_group.is_prefill() is True
86
- seq_group.update_num_computed_tokens(5)
87
- assert seq_group.is_prefill() is True
88
- seq_group.update_num_computed_tokens(1)
89
- assert seq_group.is_prefill() is False
90
- seqs = seq_group.get_seqs()
91
- assert len(seqs) == 1
92
- seqs[0].data.append_token_id(1, logprob=0.0)
93
- for seq in seq_group.get_seqs():
94
- seq.reset_state_for_recompute()
95
- assert seq_group.is_prefill() is True
96
- seq_group.update_num_computed_tokens(5)
97
- assert seq_group.is_prefill() is True
98
- seq_group.update_num_computed_tokens(7)
99
- assert seq_group.is_prefill() is True
100
- seq_group.update_num_computed_tokens(1)
101
- assert seq_group.is_prefill() is False
6
+from vllm.sequence import IntermediateTensors
7
8
9
def test_sequence_intermediate_tensors_equal():