MCPcopy Create free account
hub / github.com/PaddlePaddle/FastDeploy / BenchmarkMetrics

Class BenchmarkMetrics

benchmarks/quick_benchmark.py:51–107  ·  view source on GitHub ↗

Class containing all metrics that are used in this script

Source from the content-addressed store, hash-verified

49
50@dataclass
51class BenchmarkMetrics:
52 """Class containing all metrics that are used in this script"""
53
54 completed: int
55 total_input: int
56 total_output: int
57 request_throughput: float
58 request_goodput: float
59 output_throughput: float
60 total_token_throughput: float
61 mean_s_decode: float
62 median_s_decode: float
63 std_s_decode: float
64 percentiles_s_decode: list[tuple[float, float]]
65 mean_ttft_ms: float
66 median_ttft_ms: float
67 std_ttft_ms: float
68 percentiles_ttft_ms: list[tuple[float, float]]
69 mean_s_ttft_ms: float
70 median_s_ttft_ms: float
71 std_s_ttft_ms: float
72 percentiles_s_ttft_ms: list[tuple[float, float]]
73 mean_tpot_ms: float
74 median_tpot_ms: float
75 std_tpot_ms: float
76 percentiles_tpot_ms: list[tuple[float, float]]
77 mean_itl_ms: float
78 median_itl_ms: float
79 std_itl_ms: float
80 percentiles_itl_ms: list[tuple[float, float]]
81 mean_s_itl_ms: float
82 median_s_itl_ms: float
83 std_s_itl_ms: float
84 percentiles_s_itl_ms: list[tuple[float, float]]
85 # E2EL stands for end-to-end latency per request.
86 # It is the time taken on the client side from sending
87 # a request to receiving a complete response.
88 mean_e2el_ms: float
89 median_e2el_ms: float
90 std_e2el_ms: float
91 percentiles_e2el_ms: list[tuple[float, float]]
92 mean_s_e2el_ms: float
93 median_s_e2el_ms: float
94 std_s_e2el_ms: float
95 percentiles_s_e2el_ms: list[tuple[float, float]]
96 mean_input_len: float
97 median_input_len: float
98 std_input_len: float
99 percentiles_input_len: list[tuple[float, float]]
100 mean_s_input_len: float
101 median_s_input_len: float
102 std_s_input_len: float
103 percentiles_s_input_len: list[tuple[float, float]]
104 mean_output_len: float
105 median_output_len: float
106 std_output_len: float
107 percentiles_output_len: list[tuple[float, float]]
108

Callers 1

calculate_metricsFunction · 0.70

Calls

no outgoing calls

Tested by

no test coverage detected