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 5, 18:30 UTC
vLLM Benchmark Suite perf regression after PR#32723 (#38576)
Signed-off-by: louie-tsai <louie.tsai@intel.com>
Louie Tsai committed
Mar 30, 2026 at 22:23 UTC
44eef0ca1ece551e5ee40e099f6f9bb67fc5733c
6 files changed
+15
-1
.buildkite/performance-benchmarks/tests/serving-tests-arm64-cpu.json
+2
-1
@@ -36,6 +36,7 @@
36
"model": "meta-llama/Llama-3.1-8B-Instruct",
37
"backend": "vllm",
38
"ignore-eos": "",
39
+ "temperature": 0,
40
"num_prompts": 200
41
}
42
},
@@ -127,4 +128,4 @@
128
}
129
}
130
]
130
-}
\ No newline at end of file
131
+}
.buildkite/performance-benchmarks/tests/serving-tests-cpu-asr.json
+1
@@ -22,6 +22,7 @@
22
"hf_split": "test",
23
"no_stream": "",
24
"no_oversample": "",
25
+ "temperature": 0,
26
"num_prompts": 200
27
}
28
},
.buildkite/performance-benchmarks/tests/serving-tests-cpu-text.json
+1
@@ -26,6 +26,7 @@
26
"model": "meta-llama/Llama-3.1-8B-Instruct",
27
"backend": "vllm",
28
"ignore-eos": "",
29
+ "temperature": 0,
30
"num_prompts": 200
31
}
32
},
.buildkite/performance-benchmarks/tests/serving-tests-cpu.json
+1
@@ -26,6 +26,7 @@
26
"model": "meta-llama/Llama-3.1-8B-Instruct",
27
"backend": "vllm",
28
"ignore-eos": "",
29
+ "temperature": 0,
30
"num_prompts": 200
31
}
32
},
.buildkite/performance-benchmarks/tests/serving-tests-hpu.json
+6
@@ -21,6 +21,7 @@
21
"backend": "vllm",
22
"dataset_name": "sharegpt",
23
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
24
+ "temperature": 0,
25
"num_prompts": 200
26
}
27
},
@@ -47,6 +48,7 @@
48
"backend": "vllm",
49
"dataset_name": "sharegpt",
50
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
51
+ "temperature": 0,
52
"num_prompts": 200
53
}
54
},
@@ -73,6 +75,7 @@
75
"backend": "vllm",
76
"dataset_name": "sharegpt",
77
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
78
+ "temperature": 0,
79
"num_prompts": 200
80
}
81
},
@@ -100,6 +103,7 @@
103
"backend": "vllm",
104
"dataset_name": "sharegpt",
105
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
106
+ "temperature": 0,
107
"num_prompts": 200
108
}
109
},
@@ -127,6 +131,7 @@
131
"backend": "vllm",
132
"dataset_name": "sharegpt",
133
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
134
+ "temperature": 0,
135
"num_prompts": 200
136
}
137
},
@@ -151,6 +156,7 @@
156
"backend": "vllm",
157
"dataset_name": "sharegpt",
158
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
159
+ "temperature": 0,
160
"num_prompts": 200
161
}
162
}
.buildkite/performance-benchmarks/tests/serving-tests.json
+4
@@ -13,6 +13,7 @@
13
"backend": "vllm",
14
"dataset_name": "sharegpt",
15
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
16
+ "temperature": 0,
17
"num_prompts": 200
18
}
19
},
@@ -30,6 +31,7 @@
31
"backend": "vllm",
32
"dataset_name": "sharegpt",
33
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
34
+ "temperature": 0,
35
"num_prompts": 200
36
}
37
},
@@ -47,6 +49,7 @@
49
"backend": "vllm",
50
"dataset_name": "sharegpt",
51
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
52
+ "temperature": 0,
53
"num_prompts": 200
54
}
55
},
@@ -67,6 +70,7 @@
70
"backend": "vllm",
71
"dataset_name": "sharegpt",
72
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
73
+ "temperature": 0,
74
"num_prompts": 200
75
}
76
}