look up multi_tp key (#3699)

### What this PR does / why we need it?
In multi-Tensor Parallel (TP) scenarios, the KV pool only queries the
first GPU card. When keys on other cards are released, the query result
still returns as successful, introducing accuracy issues. This PR
modifies the KV pool's query logic to check all cards, resolving this
problem.
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

---------

Signed-off-by: fems14 <1804143737@qq.com>
This commit is contained in:
fems14
2025-10-24 17:23:36 +08:00
committed by GitHub
parent c83efcb9e4
commit 82a4970fe9
3 changed files with 95 additions and 22 deletions

View File

@@ -68,7 +68,7 @@ class Mooncakestore():
def exists(self, key: MooncakeEngineKey) -> bool:
return self.store.is_exist(key.to_string()) == 1
def batch_exists(self, keys: list[str]) -> list[bool]:
def batch_exists(self, keys: list[str]) -> list[int]:
return self.store.batch_is_exist(keys)
def register_buffer(self, ptr, length):