Skip to content

GitLab

"vllm/model_executor/layers/attention" did not exist on "925f3332cac488e5ad2dbc8f5c6d5f42d2556816"
Switch branch/tag
Find file Normal viewHistoryPermalink
test_profile_sla_dryrun.py 3.93 KB
Newer Older
Hongkuan Zhou's avatar
test: add test for pre-deployment script (#2857)  
Hongkuan Zhou committed
1
2
3
4
5
6
7 
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

"""
Test suite for profile_sla dry-run functionality.

This test ensures that the profile_sla script can successfully run in dry-run mode
Hongkuan Zhou's avatar
feat: support trtllm in sla-planner (#2980)  
Hongkuan Zhou committed
8 
for vllm, sglang, and trtllm backends with their respective disagg.yaml configurations.
Hongkuan Zhou's avatar
test: add test for pre-deployment script (#2857)  
Hongkuan Zhou committed
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 
"""

import sys
from pathlib import Path

import pytest

# Add the project root to sys.path to enable imports
project_root = Path(__file__).parent.parent.parent
sys.path.insert(0, str(project_root))

from benchmarks.profiler.profile_sla import run_profile  # noqa: E402


class TestProfileSLADryRun:
    """Test class for profile_sla dry-run functionality."""

    @pytest.fixture
    def vllm_args(self):
        """Create arguments for vllm backend dry-run test."""

        class Args:
            backend = "vllm"
            config = "components/backends/vllm/deploy/disagg.yaml"
            output_dir = "/tmp/test_profiling_results"
            namespace = "test-namespace"
            min_num_gpus_per_engine = 1
            max_num_gpus_per_engine = 8
            skip_existing_results = False
            force_rerun = False
            isl = 3000
            osl = 500
            ttft = 50
            itl = 10
            max_context_length = 16384
            prefill_interpolation_granularity = 16
            decode_interpolation_granularity = 6
            service_name = ""
            dry_run = True

        return Args()

    @pytest.fixture
    def sglang_args(self):
        """Create arguments for sglang backend dry-run test."""

        class Args:
            backend = "sglang"
            config = "components/backends/sglang/deploy/disagg.yaml"
            output_dir = "/tmp/test_profiling_results"
            namespace = "test-namespace"
            min_num_gpus_per_engine = 1
            max_num_gpus_per_engine = 8
            skip_existing_results = False
            force_rerun = False
            isl = 3000
            osl = 500
            ttft = 50
            itl = 10
            max_context_length = 16384
            prefill_interpolation_granularity = 16
            decode_interpolation_granularity = 6
            service_name = ""
            dry_run = True

        return Args()

    @pytest.mark.pre_merge
    @pytest.mark.asyncio
    async def test_vllm_dryrun(self, vllm_args):
        """Test that profile_sla dry-run works for vllm backend with disagg.yaml config."""
        # Run the profile in dry-run mode - should complete without errors
        await run_profile(vllm_args)

    @pytest.mark.pre_merge
    @pytest.mark.asyncio
    async def test_sglang_dryrun(self, sglang_args):
        """Test that profile_sla dry-run works for sglang backend with disagg.yaml config."""
        # Run the profile in dry-run mode - should complete without errors
        await run_profile(sglang_args)
Hongkuan Zhou's avatar
feat: support trtllm in sla-planner (#2980)  
Hongkuan Zhou committed
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 

    @pytest.fixture
    def trtllm_args(self):
        """Create arguments for trtllm backend dry-run test."""

        class Args:
            backend = "trtllm"
            config = "components/backends/trtllm/deploy/disagg.yaml"
            output_dir = "/tmp/test_profiling_results"
            namespace = "test-namespace"
            min_num_gpus_per_engine = 1
            max_num_gpus_per_engine = 8
            skip_existing_results = False
            force_rerun = False
            isl = 3000
            osl = 500
            ttft = 50
            itl = 10
            max_context_length = 16384
            prefill_interpolation_granularity = 16
            decode_interpolation_granularity = 6
            service_name = ""
            dry_run = True

        return Args()

    @pytest.mark.pre_merge
    @pytest.mark.asyncio
    async def test_trtllm_dryrun(self, trtllm_args):
        """Test that profile_sla dry-run works for trtllm backend with disagg.yaml config."""
        # Run the profile in dry-run mode - should complete without errors
        await run_profile(trtllm_args)