Convert formatting to use ruff instead of yapf + isort (#26247)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
+42
-31
@@ -8,8 +8,12 @@ import torch
|
||||
|
||||
from vllm.config import VllmConfig
|
||||
from vllm.logger import init_logger
|
||||
from vllm.lora.models import (LoRAModel, LoRAModelManager,
|
||||
LRUCacheLoRAModelManager, create_lora_manager)
|
||||
from vllm.lora.models import (
|
||||
LoRAModel,
|
||||
LoRAModelManager,
|
||||
LRUCacheLoRAModelManager,
|
||||
create_lora_manager,
|
||||
)
|
||||
from vllm.lora.peft_helper import PEFTHelper
|
||||
from vllm.lora.request import LoRARequest
|
||||
from vllm.lora.utils import get_adapter_absolute_path
|
||||
@@ -39,7 +43,8 @@ class WorkerLoRAManager:
|
||||
self._cached_dummy_lora: Union[None, Literal[False], LoRAModel] = False
|
||||
self.max_num_seqs = vllm_config.scheduler_config.max_num_seqs
|
||||
self.max_num_batched_tokens = (
|
||||
vllm_config.scheduler_config.max_num_batched_tokens)
|
||||
vllm_config.scheduler_config.max_num_batched_tokens
|
||||
)
|
||||
self.vocab_size = vllm_config.model_config.get_vocab_size()
|
||||
self.lora_config = vllm_config.lora_config
|
||||
|
||||
@@ -81,15 +86,12 @@ class WorkerLoRAManager:
|
||||
|
||||
def _load_adapter(self, lora_request: LoRARequest) -> LoRAModel:
|
||||
try:
|
||||
supported_lora_modules = (
|
||||
self._adapter_manager.supported_lora_modules)
|
||||
packed_modules_mapping = (
|
||||
self._adapter_manager.packed_modules_mapping)
|
||||
supported_lora_modules = self._adapter_manager.supported_lora_modules
|
||||
packed_modules_mapping = self._adapter_manager.packed_modules_mapping
|
||||
expected_lora_modules: list[str] = []
|
||||
for module in supported_lora_modules:
|
||||
if module in packed_modules_mapping:
|
||||
expected_lora_modules.extend(
|
||||
packed_modules_mapping[module])
|
||||
expected_lora_modules.extend(packed_modules_mapping[module])
|
||||
else:
|
||||
expected_lora_modules.append(module)
|
||||
|
||||
@@ -97,8 +99,10 @@ class WorkerLoRAManager:
|
||||
lora_path = get_adapter_absolute_path(lora_request.lora_path)
|
||||
|
||||
peft_helper = PEFTHelper.from_local_dir(
|
||||
lora_path, self.max_position_embeddings,
|
||||
lora_request.tensorizer_config_dict)
|
||||
lora_path,
|
||||
self.max_position_embeddings,
|
||||
lora_request.tensorizer_config_dict,
|
||||
)
|
||||
|
||||
# Validates the LoRA configuration against requirements before
|
||||
# loading weights, throwing an exception if validation fails.
|
||||
@@ -116,12 +120,13 @@ class WorkerLoRAManager:
|
||||
lora_model_id=lora_request.lora_int_id,
|
||||
device="cpu",
|
||||
dtype=self.lora_config.lora_dtype,
|
||||
target_embedding_padding=self.vocab_size +
|
||||
self.lora_config.lora_extra_vocab_size,
|
||||
target_embedding_padding=self.vocab_size
|
||||
+ self.lora_config.lora_extra_vocab_size,
|
||||
embedding_modules=self.embedding_modules,
|
||||
embedding_padding_modules=self.embedding_padding_modules,
|
||||
tensorizer_config_dict=lora_request.tensorizer_config_dict,
|
||||
weights_mapper=hf_to_vllm_mapper)
|
||||
weights_mapper=hf_to_vllm_mapper,
|
||||
)
|
||||
|
||||
except FileNotFoundError as e:
|
||||
# FileNotFoundError should be raised if both
|
||||
@@ -131,26 +136,29 @@ class WorkerLoRAManager:
|
||||
# For NotFoundError
|
||||
raise ValueError(
|
||||
f"Loading lora {lora_request.lora_name} failed: No adapter "
|
||||
f"found for {lora_request.lora_path}") from e
|
||||
f"found for {lora_request.lora_path}"
|
||||
) from e
|
||||
except Exception as e:
|
||||
# For BadRequestError
|
||||
raise e
|
||||
|
||||
if lora.extra_vocab_size > self.lora_config.lora_extra_vocab_size:
|
||||
raise ValueError(f"LoRA added vocab size {lora.extra_vocab_size} "
|
||||
f"is greater than lora_extra_vocab_size "
|
||||
f"{self.lora_config.lora_extra_vocab_size}.")
|
||||
raise ValueError(
|
||||
f"LoRA added vocab size {lora.extra_vocab_size} "
|
||||
f"is greater than lora_extra_vocab_size "
|
||||
f"{self.lora_config.lora_extra_vocab_size}."
|
||||
)
|
||||
return lora
|
||||
|
||||
def add_dummy_lora(self, lora_request: LoRARequest, rank: int) -> bool:
|
||||
if lora_request.lora_int_id in self.list_adapters():
|
||||
return False
|
||||
if isinstance(self._cached_dummy_lora, LoRAModel):
|
||||
dummy_lora = self._cached_dummy_lora.clone(
|
||||
lora_request.lora_int_id)
|
||||
dummy_lora = self._cached_dummy_lora.clone(lora_request.lora_int_id)
|
||||
else:
|
||||
dummy_lora = self._adapter_manager.create_dummy_lora(
|
||||
lora_request.lora_int_id, rank, self.embedding_modules)
|
||||
lora_request.lora_int_id, rank, self.embedding_modules
|
||||
)
|
||||
if self._cached_dummy_lora is None:
|
||||
self._cached_dummy_lora = dummy_lora
|
||||
return self._adapter_manager.add_adapter(dummy_lora)
|
||||
@@ -158,8 +166,7 @@ class WorkerLoRAManager:
|
||||
def pin_adapter(self, adapter_id: int) -> bool:
|
||||
return self._adapter_manager.pin_adapter(adapter_id)
|
||||
|
||||
def set_active_adapters(self, requests: set[Any],
|
||||
mapping: Optional[Any]) -> None:
|
||||
def set_active_adapters(self, requests: set[Any], mapping: Optional[Any]) -> None:
|
||||
self._apply_adapters(requests)
|
||||
if mapping is not None:
|
||||
self._adapter_manager.set_adapter_mapping(mapping)
|
||||
@@ -168,13 +175,15 @@ class WorkerLoRAManager:
|
||||
existing_adapters = self.list_adapters()
|
||||
models_map = {
|
||||
adapter_request.adapter_id: adapter_request
|
||||
for adapter_request in adapter_requests if adapter_request
|
||||
for adapter_request in adapter_requests
|
||||
if adapter_request
|
||||
}
|
||||
if len(models_map) > self._adapter_manager.adapter_slots:
|
||||
raise RuntimeError(
|
||||
f"Number of requested models ({len(models_map)}) is greater "
|
||||
"than the number of GPU model slots "
|
||||
f"({self._adapter_manager.adapter_slots}).")
|
||||
f"({self._adapter_manager.adapter_slots})."
|
||||
)
|
||||
requested_ids = set(models_map)
|
||||
for adapter_id in existing_adapters - requested_ids:
|
||||
self.remove_adapter(adapter_id)
|
||||
@@ -227,13 +236,15 @@ class LRUCacheWorkerLoRAManager(WorkerLoRAManager):
|
||||
def _apply_adapters(self, lora_requests: set[LoRARequest]) -> None:
|
||||
loras_map = {
|
||||
lora_request.lora_int_id: lora_request
|
||||
for lora_request in lora_requests if lora_request
|
||||
for lora_request in lora_requests
|
||||
if lora_request
|
||||
}
|
||||
if len(loras_map) > self._adapter_manager.lora_slots:
|
||||
raise RuntimeError(
|
||||
f"Number of requested LoRAs ({len(loras_map)}) is greater "
|
||||
"than the number of GPU LoRA slots "
|
||||
f"({self._adapter_manager.lora_slots}).")
|
||||
f"({self._adapter_manager.lora_slots})."
|
||||
)
|
||||
for lora in loras_map.values():
|
||||
self.add_adapter(lora)
|
||||
|
||||
@@ -253,15 +264,15 @@ class LRUCacheWorkerLoRAManager(WorkerLoRAManager):
|
||||
# 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:
|
||||
assert isinstance(self._adapter_manager,
|
||||
LRUCacheLoRAModelManager)
|
||||
assert isinstance(self._adapter_manager, LRUCacheLoRAModelManager)
|
||||
self._adapter_manager.remove_oldest_adapter()
|
||||
# Then add the new adapter to the cache
|
||||
loaded = self._adapter_manager.add_adapter(lora)
|
||||
else:
|
||||
# If the lora is already loaded, just touch it to
|
||||
# update its position in the caches
|
||||
loaded = self._adapter_manager.get_adapter(
|
||||
lora_request.lora_int_id) is not None
|
||||
loaded = (
|
||||
self._adapter_manager.get_adapter(lora_request.lora_int_id) is not None
|
||||
)
|
||||
self._adapter_manager.activate_adapter(lora_request.lora_int_id)
|
||||
return loaded
|
||||
|
||||
Reference in New Issue
Block a user