xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-15 18:35:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Kernel] Add back batch size 1536 and 3072 to MoE tuning (#5242)

Browse Source
This commit is contained in:
Woosuk Kwon 2024-06-04 09:58:47 -07:00 committed by GitHub
parent 87d5abef75
commit 27208be66e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
benchmarks/kernels/benchmark_moe.py
View File

@ -254,7 +254,9 @@ def main(args: argparse.Namespace):
use_fp8 = args.dtype == "fp8" use_fp8 = args.dtype == "fp8"
if args.batch_size is None: if args.batch_size is None:
batch_sizes = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096] batch_sizes = [
1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 1536, 2048, 3072, 4096
]
else: else:
batch_sizes = [args.batch_size] batch_sizes = [args.batch_size]