|
| 1 | +# Copyright 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved. |
| 2 | +# |
| 3 | +# Redistribution and use in source and binary forms, with or without |
| 4 | +# modification, are permitted provided that the following conditions |
| 5 | +# are met: |
| 6 | +# * Redistributions of source code must retain the above copyright |
| 7 | +# notice, this list of conditions and the following disclaimer. |
| 8 | +# * Redistributions in binary form must reproduce the above copyright |
| 9 | +# notice, this list of conditions and the following disclaimer in the |
| 10 | +# documentation and/or other materials provided with the distribution. |
| 11 | +# * Neither the name of NVIDIA CORPORATION nor the names of its |
| 12 | +# contributors may be used to endorse or promote products derived |
| 13 | +# from this software without specific prior written permission. |
| 14 | +# |
| 15 | +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY |
| 16 | +# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 17 | +# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| 18 | +# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR |
| 19 | +# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
| 20 | +# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
| 21 | +# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
| 22 | +# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY |
| 23 | +# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 24 | +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 25 | +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 26 | + |
| 27 | +import os |
| 28 | +import re |
| 29 | +import sys |
| 30 | +import unittest |
| 31 | +from functools import partial |
| 32 | + |
| 33 | +import requests |
| 34 | +import tritonclient.grpc as grpcclient |
| 35 | +from tritonclient.utils import * |
| 36 | + |
| 37 | +sys.path.append("../../common") |
| 38 | +from test_util import TestResultCollector, UserData, callback, create_vllm_request |
| 39 | + |
| 40 | + |
| 41 | +class VLLMTritonMetricsTest(TestResultCollector): |
| 42 | + def setUp(self): |
| 43 | + self.triton_client = grpcclient.InferenceServerClient(url="localhost:8001") |
| 44 | + self.tritonserver_ipaddr = os.environ.get("TRITONSERVER_IPADDR", "localhost") |
| 45 | + self.vllm_model_name = "vllm_opt" |
| 46 | + self.prompts = [ |
| 47 | + "The most dangerous animal is", |
| 48 | + "The capital of France is", |
| 49 | + "The future of AI is", |
| 50 | + ] |
| 51 | + self.sampling_parameters = {"temperature": "0", "top_p": "1"} |
| 52 | + |
| 53 | + def get_metrics(self): |
| 54 | + """ |
| 55 | + Store vllm metrics in a dictionary. |
| 56 | + """ |
| 57 | + r = requests.get(f"http://{self.tritonserver_ipaddr}:8002/metrics") |
| 58 | + r.raise_for_status() |
| 59 | + |
| 60 | + # Regular expression to match the pattern |
| 61 | + pattern = r"^(vllm:[^ {]+)(?:{.*})? ([0-9.-]+)$" |
| 62 | + vllm_dict = {} |
| 63 | + |
| 64 | + # Find all matches in the text |
| 65 | + matches = re.findall(pattern, r.text, re.MULTILINE) |
| 66 | + |
| 67 | + for match in matches: |
| 68 | + key, value = match |
| 69 | + vllm_dict[key] = float(value) if "." in value else int(value) |
| 70 | + |
| 71 | + return vllm_dict |
| 72 | + |
| 73 | + def vllm_async_stream_infer( |
| 74 | + self, |
| 75 | + prompts, |
| 76 | + sampling_parameters, |
| 77 | + stream, |
| 78 | + send_parameters_as_tensor, |
| 79 | + model_name, |
| 80 | + ): |
| 81 | + """ |
| 82 | + Helper function to send async stream infer requests to vLLM. |
| 83 | + """ |
| 84 | + user_data = UserData() |
| 85 | + number_of_vllm_reqs = len(prompts) |
| 86 | + |
| 87 | + self.triton_client.start_stream(callback=partial(callback, user_data)) |
| 88 | + for i in range(number_of_vllm_reqs): |
| 89 | + request_data = create_vllm_request( |
| 90 | + prompts[i], |
| 91 | + i, |
| 92 | + stream, |
| 93 | + sampling_parameters, |
| 94 | + model_name, |
| 95 | + send_parameters_as_tensor, |
| 96 | + ) |
| 97 | + self.triton_client.async_stream_infer( |
| 98 | + model_name=model_name, |
| 99 | + request_id=request_data["request_id"], |
| 100 | + inputs=request_data["inputs"], |
| 101 | + outputs=request_data["outputs"], |
| 102 | + parameters=sampling_parameters, |
| 103 | + ) |
| 104 | + |
| 105 | + for _ in range(number_of_vllm_reqs): |
| 106 | + result = user_data._completed_requests.get() |
| 107 | + if type(result) is InferenceServerException: |
| 108 | + print(result.message()) |
| 109 | + self.assertIsNot(type(result), InferenceServerException, str(result)) |
| 110 | + |
| 111 | + output = result.as_numpy("text_output") |
| 112 | + self.assertIsNotNone(output, "`text_output` should not be None") |
| 113 | + |
| 114 | + self.triton_client.stop_stream() |
| 115 | + |
| 116 | + def test_vllm_metrics(self): |
| 117 | + # Test vLLM metrics |
| 118 | + self.vllm_async_stream_infer( |
| 119 | + prompts=self.prompts, |
| 120 | + sampling_parameters=self.sampling_parameters, |
| 121 | + stream=False, |
| 122 | + send_parameters_as_tensor=True, |
| 123 | + model_name=self.vllm_model_name, |
| 124 | + ) |
| 125 | + metrics_dict = self.get_metrics() |
| 126 | + |
| 127 | + self.assertEqual(metrics_dict["vllm:time_to_first_token_seconds_count"], 3) |
| 128 | + self.assertTrue( |
| 129 | + 0.0001 < metrics_dict["vllm:time_to_first_token_seconds_sum"] < 0.0003 |
| 130 | + ) |
| 131 | + self.assertEqual(metrics_dict["vllm:time_per_output_token_seconds_count"], 45) |
| 132 | + self.assertTrue( |
| 133 | + 0.001 <= metrics_dict["vllm:time_per_output_token_seconds_sum"] <= 0.003 |
| 134 | + ) |
| 135 | + |
| 136 | + def tearDown(self): |
| 137 | + self.triton_client.close() |
| 138 | + |
| 139 | + |
| 140 | +if __name__ == "__main__": |
| 141 | + unittest.main() |
0 commit comments