apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Feature] Support block scheduler v1 for FD (#2928)

Browse Source 
* Support FD block scheduler v1

* Support FD block scheduler v1

* Support FD block scheduler v1

* Fix according to copilot review

* Fix according to review

* Remove is_dummy

* Fix bug when real_bsz=1

* Fix infer first token cost time

---------

Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com>
This commit is contained in:
chenjian
2025-07-23 20:31:31 +08:00
committed by GitHub
parent ca0f71bd39
commit 85a78d695d
16 changed files with 898 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
fastdeploy/cache_manager/prefix_cache_manager.py
View File

@@ -210,8 +210,16 @@ class PrefixCacheManager:
update cache config
"""
self.cache_config = cache_config
self.num_gpu_blocks = cache_config.prefill_kvcache_block_num
self.gpu_free_block_list = list(range(self.num_gpu_blocks - 1, -1, -1)) # 服务端管理的GPU上剩余的block id
if envs.ENABLE_V1_KVCACHE_SCHEDULER:
self.num_gpu_blocks = cache_config.total_block_num
self.gpu_free_block_list = list(
range(self.num_gpu_blocks - 1, -1, -1)
) # All gpu blocks are managed by cache manager
else:
self.num_gpu_blocks = cache_config.prefill_kvcache_block_num
self.gpu_free_block_list = list(
range(self.num_gpu_blocks - 1, -1, -1)
) # Only block table divided for prefill managed by server
heapq.heapify(self.gpu_free_block_list)
self.node_id_pool = list(range(self.num_gpu_blocks + self.num_cpu_blocks))
@@ -231,6 +239,15 @@ class PrefixCacheManager:
self.transfer_recv_thread = threading.Thread(target=self.recv_data_transfer_result)
self.transfer_recv_thread.start()
def can_allocate_gpu_blocks(self, num_blocks: int):
"""
Check if num_blocks gpu blocks can be allocated.
"""
if len(self.gpu_free_block_list) < num_blocks:
return False
else:
return True
def allocate_gpu_blocks(self, num_blocks):
"""
allocate gpu blocks.