[Feature] Support KV cache offloading and disagg prefill with LMCache connector. (#12953)

This commit is contained in:
Jiayi Yao
2025-02-25 00:38:42 -08:00
committed by GitHub
parent 3173c3b34e
commit 2f42a4888c
5 changed files with 310 additions and 2 deletions
+2 -2
View File
@@ -962,8 +962,8 @@ def ensure_kv_transfer_initialized(vllm_config: "VllmConfig") -> None:
return
if all([
vllm_config.kv_transfer_config.need_kv_parallel_group, _KV_TRANSFER
is None
vllm_config.kv_transfer_config.is_kv_transfer_instance,
_KV_TRANSFER is None
]):
_KV_TRANSFER = kv_transfer.KVTransferAgent(
rank=get_world_group().rank,