xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-23 02:52:22 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge 0718a782eea404a90f461fed3da6eb48f30f5114 into 254f6b986720c92ddf97fbb1a6a6465da8e87e29

Browse Source
This commit is contained in:
Jackmin801 2025-12-25 00:22:44 +00:00 committed by GitHub
commit 0562ae7cd3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 12 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
vllm/entrypoints/openai/serving_models.py
View File

@ -142,9 +142,13 @@ class OpenAIServingModels:
return error_check_ret
lora_path = request.lora_path
unique_id = self.lora_id_counter.inc(1)
lora_int_id = (
self.lora_requests[lora_name].lora_int_id
if lora_name in self.lora_requests
else self.lora_id_counter.inc(1)
)
lora_request = LoRARequest(
lora_name=lora_name, lora_int_id=unique_id, lora_path=lora_path
lora_name=lora_name, lora_int_id=lora_int_id, lora_path=lora_path
)
if base_model_name is not None and self.is_base_model(base_model_name):
lora_request.base_model_name = base_model_name
@ -197,15 +201,6 @@ class OpenAIServingModels:
status_code=HTTPStatus.BAD_REQUEST,
)
# Check if the lora adapter with the given name already exists
if request.lora_name in self.lora_requests:
return create_error_response(
message=f"The lora adapter '{request.lora_name}' has already been "
"loaded.",
err_type="InvalidUserInput",
status_code=HTTPStatus.BAD_REQUEST,
)
return None
async def _check_unload_lora_adapter_request(

9
vllm/lora/worker_manager.py
View File

@ -236,21 +236,24 @@ class LRUCacheWorkerLoRAManager(WorkerLoRAManager):
f"({self._adapter_manager.lora_slots})."
)
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
# times for the same adapter when using multiple API servers.
# This is ok because it's currently only called from
# 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
# evicting any existing adapters.
# This may cause the # of loaded lora adapters to very temporarily
# exceed `--max-cpu-loras`.
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
# evict if the oldest adapter if so
if len(self._adapter_manager) + 1 > self._adapter_manager.capacity: