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 8, 22:45 UTC
Save startup benchmark results as a list of values (#33629)
Signed-off-by: Huy Do <huydhn@gmail.com>
Huy Do committed
Feb 3, 2026 at 20:37 UTC
2647163674720242cee78dae3f7c7b98539ed029
2 files changed
+10
-4
vllm/benchmarks/lib/utils.py
+6
@@ -47,6 +47,12 @@ def convert_to_pytorch_benchmark_format(
47
return records
48
49
for name, benchmark_values in metrics.items():
50
+ if not isinstance(benchmark_values, list):
51
+ raise TypeError(
52
+ f"benchmark_values for metric '{name}' must be a list, "
53
+ f"but got {type(benchmark_values).__name__}"
54
+ )
55
+
56
record = {
57
"benchmark": {
58
"name": "vLLM benchmark",
vllm/benchmarks/startup.py
+4
-4
@@ -101,7 +101,7 @@ def save_to_pytorch_benchmark_format(
101
cold_startup_records = convert_to_pytorch_benchmark_format(
102
args=args,
103
metrics={
104
- "avg_cold_startup_time": results["avg_cold_startup_time"],
104
+ "avg_cold_startup_time": [results["avg_cold_startup_time"]],
105
},
106
extra_info={
107
"cold_startup_times": results["cold_startup_times"],
@@ -114,7 +114,7 @@ def save_to_pytorch_benchmark_format(
114
cold_compilation_records = convert_to_pytorch_benchmark_format(
115
args=args,
116
metrics={
117
- "avg_cold_compilation_time": results["avg_cold_compilation_time"],
117
+ "avg_cold_compilation_time": [results["avg_cold_compilation_time"]],
118
},
119
extra_info={
120
"cold_compilation_times": results["cold_compilation_times"],
@@ -129,7 +129,7 @@ def save_to_pytorch_benchmark_format(
129
warm_startup_records = convert_to_pytorch_benchmark_format(
130
args=args,
131
metrics={
132
- "avg_warm_startup_time": results["avg_warm_startup_time"],
132
+ "avg_warm_startup_time": [results["avg_warm_startup_time"]],
133
},
134
extra_info={
135
"warm_startup_times": results["warm_startup_times"],
@@ -142,7 +142,7 @@ def save_to_pytorch_benchmark_format(
142
warm_compilation_records = convert_to_pytorch_benchmark_format(
143
args=args,
144
metrics={
145
- "avg_warm_compilation_time": results["avg_warm_compilation_time"],
145
+ "avg_warm_compilation_time": [results["avg_warm_compilation_time"]],
146
},
147
extra_info={
148
"warm_compilation_times": results["warm_compilation_times"],