xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-04-02 18:37:07 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add assert in init of SinkFullAttentionManager

Browse Source 
Signed-off-by: yuantao <2422264527@qq.com>
This commit is contained in:
yuantao 2025-12-13 16:32:56 +08:00
parent 2470548aaa
commit a0563e7368
1 changed files with 3 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/v1/core/single_type_kv_cache_manager.py
View File

@ -797,12 +797,9 @@ class SinkFullAttentionManager(FullAttentionManager):
kv_cache_spec, block_pool, kv_cache_group_id, dcp_world_size, pcp_world_size
)
sink_len = kv_cache_spec.sink_len
if sink_len > 0:
assert sink_len % self.block_size == 0
num_sink_block = sink_len // self.block_size
self.sink_blocks = self.block_pool.free_block_queue.popleft_n(
num_sink_block
)
assert sink_len is not None and sink_len > 0 and sink_len % self.block_size == 0
num_sink_block = sink_len // self.block_size
self.sink_blocks = self.block_pool.free_block_queue.popleft_n(num_sink_block)
spec_manager_map: dict[type[KVCacheSpec], type[SingleTypeKVCacheManager]] = {