biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Convert benchmarks to ruff format (#18068)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-05-13 14:43:29 +01:00
committed by GitHub
parent b922c2ebd2
commit 009d9e7590
41 changed files with 3980 additions and 2938 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
benchmarks/kernels/benchmark_quant.py
View File

@@ -10,15 +10,17 @@ from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, FlexibleArgumentParser
@torch.inference_mode()
def main(num_tokens: int,
hidden_size: int,
static_scale: bool,
quant_dtype: torch.dtype,
dtype: torch.dtype,
seed: int = 0,
do_profile: bool = False,
num_warmup_iters: int = 5,
num_iters: int = 100) -> None:
def main(
num_tokens: int,
hidden_size: int,
static_scale: bool,
quant_dtype: torch.dtype,
dtype: torch.dtype,
seed: int = 0,
do_profile: bool = False,
num_warmup_iters: int = 5,
num_iters: int = 100,
) -> None:
current_platform.seed_everything(seed)
torch.set_default_device("cuda")
@@ -56,7 +58,7 @@ def main(num_tokens: int,
print(f"Kernel running time: {latency * 1000000:.3f} us")
if __name__ == '__main__':
if __name__ == "__main__":
def to_torch_dtype(dt):
if dt == "int8":
@@ -66,37 +68,40 @@ if __name__ == '__main__':
raise ValueError(f"Unsupported dtype: {dt}")
parser = FlexibleArgumentParser(
description="Benchmark the quantization (fp8 or int8) kernel.")
description="Benchmark the quantization (fp8 or int8) kernel."
)
parser.add_argument("--num-tokens", type=int, default=4096)
parser.add_argument("--hidden-size", type=int, default=8192)
parser.add_argument("--static-scale", action="store_true")
parser.add_argument("--quant-dtype",
type=str,
choices=["fp8", "int8"],
default="int8")
parser.add_argument("--dtype",
type=str,
choices=["half", "bfloat16", "float"],
default="half")
parser.add_argument(
"--quant-dtype", type=str, choices=["fp8", "int8"], default="int8"
)
parser.add_argument(
"--dtype", type=str, choices=["half", "bfloat16", "float"], default="half"
)
parser.add_argument("--seed", type=int, default=0)
parser.add_argument("--profile", action="store_true")
parser.add_argument("--num-warmup-iters", type=int, default=5)
parser.add_argument("--num-iters",
type=int,
default=100,
help="Number of benchmark iterations. "
"If --profile is set, this number is ignored")
parser.add_argument(
"--num-iters",
type=int,
default=100,
help="Number of benchmark iterations. "
"If --profile is set, this number is ignored",
)
args = parser.parse_args()
print(args)
main(num_tokens=args.num_tokens,
hidden_size=args.hidden_size,
static_scale=args.static_scale,
quant_dtype=to_torch_dtype(args.quant_dtype),
dtype=STR_DTYPE_TO_TORCH_DTYPE[args.dtype],
seed=args.seed,
do_profile=args.profile,
num_warmup_iters=args.num_warmup_iters,
num_iters=args.num_iters)
main(
num_tokens=args.num_tokens,
hidden_size=args.hidden_size,
static_scale=args.static_scale,
quant_dtype=to_torch_dtype(args.quant_dtype),
dtype=STR_DTYPE_TO_TORCH_DTYPE[args.dtype],
seed=args.seed,
do_profile=args.profile,
num_warmup_iters=args.num_warmup_iters,
num_iters=args.num_iters,
)