xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-06 09:42:13 +08:00
Code Issues Packages Projects Releases Wiki Activity

make LRUCacheWorkerLoRAManager force reload lora when not doing _apply_adapters

Browse Source 
Signed-off-by: Jackmin801 <ongjackm@gmail.com>
This commit is contained in:
Jackmin801 2025-12-24 23:56:52 +00:00
parent 3cce5bd947
commit 0616d19f8b
1 changed files with 6 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/lora/worker_manager.py
View File

@ -236,21 +236,24 @@ class LRUCacheWorkerLoRAManager(WorkerLoRAManager):
f"({self._adapter_manager.lora_slots})." f"({self._adapter_manager.lora_slots})."
) )
for lora in loras_map.values(): for lora in loras_map.values():
self.add_adapter(lora) self.add_adapter(lora, force_load=False)
def add_adapter(self, lora_request: LoRARequest) -> bool: def add_adapter(self, lora_request: LoRARequest, force_load: bool = True) -> bool:
# Note that this method is not thread-safe. It may be invoked multiple # Note that this method is not thread-safe. It may be invoked multiple
# times for the same adapter when using multiple API servers. # times for the same adapter when using multiple API servers.
# This is ok because it's currently only called from # This is ok because it's currently only called from
# the single-threaded core engine loop. # the single-threaded core engine loop.
if lora_request.lora_int_id not in self.list_adapters(): if lora_request.lora_int_id not in self.list_adapters() or force_load:
# Load the new adapter first to ensure it is actually valid, before # Load the new adapter first to ensure it is actually valid, before
# evicting any existing adapters. # evicting any existing adapters.
# This may cause the # of loaded lora adapters to very temporarily # This may cause the # of loaded lora adapters to very temporarily
# exceed `--max-cpu-loras`. # exceed `--max-cpu-loras`.
lora = self._load_adapter(lora_request) lora = self._load_adapter(lora_request)
# Remove the existing adapter if it exists
self._adapter_manager.remove_adapter(lora.id)
# Loading succeeded, now check if we will exceed cache capacity and # Loading succeeded, now check if we will exceed cache capacity and
# evict if the oldest adapter if so # evict if the oldest adapter if so
if len(self._adapter_manager) + 1 > self._adapter_manager.capacity: if len(self._adapter_manager) + 1 > self._adapter_manager.capacity: