xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-14 19:15:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Core][Perf] Only invoke save_new_computed_blocks when computed blocks are not empty (#27799)

Browse Source 
Signed-off-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>
This commit is contained in:
Jialin Ouyang 2025-10-30 12:47:30 -07:00 committed by GitHub
parent a8141fa649
commit 4b68c4a55b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 7 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
vllm/v1/core/kv_cache_manager.py
View File

@ -306,11 +306,12 @@ class KVCacheManager:
"Computed blocks should be empty when prefix caching is disabled" "Computed blocks should be empty when prefix caching is disabled"
) )
# Append the new computed blocks to the request blocks until now to if new_computed_block_list is not self.empty_kv_cache_blocks.blocks:
# avoid the case where the new blocks cannot be allocated. # Append the new computed blocks to the request blocks until now to
self.coordinator.save_new_computed_blocks( # avoid the case where the new blocks cannot be allocated.
request.request_id, new_computed_block_list self.coordinator.save_new_computed_blocks(
) request.request_id, new_computed_block_list
)
new_blocks = self.coordinator.allocate_new_blocks( new_blocks = self.coordinator.allocate_new_blocks(
request.request_id, num_tokens_need_slot, num_encoder_tokens request.request_id, num_tokens_need_slot, num_encoder_tokens

2
vllm/v1/core/single_type_kv_cache_manager.py
View File

@ -151,7 +151,7 @@ class SingleTypeKVCacheManager(ABC):
num_tokens: The total number of tokens that need to be cached num_tokens: The total number of tokens that need to be cached
(including tokens that are already cached). (including tokens that are already cached).
""" """
num_cached_blocks = self.num_cached_block[request.request_id] num_cached_blocks = self.num_cached_block.get(request.request_id, 0)
num_full_blocks = num_tokens // self.block_size num_full_blocks = num_tokens // self.block_size
if num_cached_blocks >= num_full_blocks: if num_cached_blocks >= num_full_blocks: