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 9, 05:00 UTC
[Misc] Replace deprecated interface seed_everything (#33474)
Signed-off-by: esmeetu <jasonailu87@gmail.com>
Roy Wang committed
Jan 31, 2026 at 21:38 UTC
68feb76a6f89abd6a52116c99b9b39e538a9a9ac
2 files changed
+4
-4
benchmarks/kernels/cpu/benchmark_cpu_attn.py
+2
-2
@@ -14,7 +14,7 @@ from vllm._custom_ops import (
14
)
15
from vllm.platforms import CpuArchEnum, current_platform
16
from vllm.utils.argparse_utils import FlexibleArgumentParser
17
-from vllm.utils.torch_utils import STR_DTYPE_TO_TORCH_DTYPE
17
+from vllm.utils.torch_utils import STR_DTYPE_TO_TORCH_DTYPE, set_random_seed
18
from vllm.v1.attention.backends.cpu_attn import CPUAttentionBackend, _get_attn_isa
19
20
@@ -58,7 +58,7 @@ def main(
58
seed: int = 0,
59
iters: int = 20,
60
) -> None:
61
- current_platform.seed_everything(seed)
61
+ set_random_seed(seed)
62
num_seqs = len(seq_lens)
63
query_lens = [x[0] for x in seq_lens]
64
kv_lens = [x[1] for x in seq_lens]
benchmarks/kernels/cpu/benchmark_cpu_fused_moe.py
+2
-2
@@ -7,8 +7,8 @@ import time
7
import numpy as np
8
import torch
9
10
-from vllm.platforms import current_platform
10
from vllm.utils.argparse_utils import FlexibleArgumentParser
11
+from vllm.utils.torch_utils import set_random_seed
12
13
# Check if CPU MoE operations are available
14
try:
@@ -41,7 +41,7 @@ def main(
41
seed: int = 0,
42
iters: int = 20,
43
) -> None:
44
- current_platform.seed_everything(seed)
44
+ set_random_seed(seed)
45
# up_dim = 2 * intermediate_size for gate + up projection
46
up_dim = 2 * intermediate_size
47