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

[Core] Make BlockSpaceManagerV2 the default BlockManager to use. (#8678)

Browse Source
This commit is contained in:
sroy745
2024-10-03 09:44:21 -07:00
committed by GitHub
parent 83caf35e08
commit f5d72b2fc6
2 changed files with 8 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/engine/arg_utils.py
View File

@@ -107,7 +107,7 @@ class EngineArgs:
block_size: int = 16
enable_prefix_caching: bool = False
disable_sliding_window: bool = False
use_v2_block_manager: bool = False
use_v2_block_manager: bool = True
swap_space: float = 4 # GiB
cpu_offload_gb: float = 0 # GiB
gpu_memory_utilization: float = 0.90
@@ -369,9 +369,12 @@ class EngineArgs:
action='store_true',
help='Disables sliding window, '
'capping to sliding window size')
parser.add_argument('--use-v2-block-manager',
action='store_true',
help='Use BlockSpaceMangerV2.')
parser.add_argument(
'--use-v2-block-manager',
default=EngineArgs.use_v2_block_manager,
action='store_true',
help='Use BlockSpaceMangerV2. By default this is set to True. '
'Set to False to use BlockSpaceManagerV1')
parser.add_argument(
'--num-lookahead-slots',
type=int,