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

[Core] Use sha256 bytes instead of BlockHash to reduce GC overhead (#23673)

Browse Source 
Signed-off-by: linzebing <linzebing1995@gmail.com>
This commit is contained in:
Zebing Lin
2025-09-09 00:34:37 -04:00
committed by GitHub
parent bba1042c6f
commit 82dfb12e52
15 changed files with 298 additions and 283 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
tests/v1/core/utils.py
View File

@ -9,6 +9,7 @@ from vllm.config import (CacheConfig, KVTransferConfig, ModelConfig,
from vllm.multimodal.inputs import (MultiModalFeatureSpec,
MultiModalKwargsItem, PlaceholderRange)
from vllm.sampling_params import SamplingParams
from vllm.utils import sha256
from vllm.v1.core.kv_cache_utils import (get_request_block_hasher,
init_none_hash)
from vllm.v1.core.sched.async_scheduler import AsyncScheduler
@ -130,10 +131,10 @@ def create_requests(
) -> list[Request]:
global _none_hash_initialized
if not _none_hash_initialized:
init_none_hash(hash)
init_none_hash(sha256)
_none_hash_initialized = True
block_hasher = get_request_block_hasher(block_size, hash)
block_hasher = get_request_block_hasher(block_size, sha256)
sampling_params = SamplingParams(ignore_eos=False,
max_tokens=max_tokens,
stop_token_ids=stop_token_ids,