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 17, 09:15 UTC
BlockAllocator -> BlockManager
Woosuk Kwon committed
Feb 12, 2023 at 08:15 UTC
46958cf941997264ff36c23c42a71d30674b61f0
1 file changed
+3
-3
cacheflow/master/block_manager.py
+3
-3
@@ -7,7 +7,7 @@ from cacheflow.sequence import SequenceStatus
7
from cacheflow.utils import Device
8
9
10
-class BlockAllocator:
10
+class BlockManager:
11
12
def __init__(
13
self,
@@ -63,8 +63,8 @@ class BlockSpaceManager:
63
self.num_total_gpu_blocks = num_gpu_blocks
64
self.num_total_cpu_blocks = num_cpu_blocks
65
66
- self.gpu_allocator = BlockAllocator(Device.GPU, block_size, num_gpu_blocks)
67
- self.cpu_allocator = BlockAllocator(Device.CPU, block_size, num_cpu_blocks)
66
+ self.gpu_allocator = BlockManager(Device.GPU, block_size, num_gpu_blocks)
67
+ self.cpu_allocator = BlockManager(Device.CPU, block_size, num_cpu_blocks)
68
69
# Mapping: seq_id -> BlockTable.
70
self.block_tables: Dict[int, BlockTable] = {}